【作者】 王素娟；

【导师】 许健民；

【作者基本信息】 中国气象科学研究院 ， 气象学， 2009， 博士

【摘要】 我国第二代静止气象卫星FY-4的三轴稳定平台能装载多种遥感载荷,实现三维探测,显著提高探测效率,并可对地球进行凝视观测;卫星拟定搭载的高精度星敏感器将用来实现高精度的卫星姿态测量,它以可精确定位的恒星系统作为绝对参照系,通过对恒星的观测、识别、计算得到星敏感器光轴的指向或三轴姿态,并根据星敏感器的安装方位等信息,计算出在轨运行的卫星的姿态。目前国产星敏感器在精度、数据更新率、预期寿命等关键性能参数都与国外先进水平有较大差距,这就给FY-4卫星的姿态控制技术带来了更大的挑战。本学位论文针对我国FY-4气象卫星三轴稳定姿态控制方式的实际需求,尝试通过研发基于地面应用系统的星图识别及姿态确定算法,通过对地面存储的星图的识别,提高出姿态率,进而得到更精细的姿态变化过程,有利于提高卫星姿态确定及图像定位的精度。本论文通过“静止气象卫星恒星定姿技术”的深入研究,确定并实现了基于星敏感器的FY-4静止气象卫星姿态确定的若干算法。首次在导航星选取研究方面引入位置精度阈值、辅助导航星等新概念,提出了一种导航星的优选方法;在星图识别算法方面,对基于三角形星图识别的全天球自主星图识别算法和局部天球星图识别算法进行了改进;在姿态反演算法研究方面,提出了理想条件下改进的TRIAD算法最佳恒星分布特征和理想条件下QUEST算法最佳星三角分布特征。本论文的主要研究工作和结论如下:（一）导航星的选取本研究给出了一种导航星的选取方法,该方法以目前公认的精度最高的J1991.25 Hipparcos星表为中心,以星敏感器光学系统的观测极限为星等阈值,联合Tycho-2星表、SAO星表及Pulkovo视向速度表,通过精密历元转换使J1991.25 Hipparcos星表中符合精密历元条件的恒星精密历元至J2000.0历元时刻,得到J2000.0 Hipparcos星表。在相同的历元J2000.0时刻,选取J2000.0 Hipparcos星表和Tycho-2星表中位置精度阈值在1角秒内的共星作为导航星,选取除导航星外的J2000.0 Hipparcos星表和SAO星表中位置精度阈值在1角秒内的共星作为辅助导航星,通过模拟星图识别失败个例分析提取冗余导航星,使其排除在导航星之外。在选取导航星的过程中,位置精度阈值的引入,使得高精度的恒星作为导航星,为高精度的卫星姿态反演奠定了基础;辅助导航星的引入,与单一使用Tycho-2星表相比引入了大量的亮星（189颗辅助导航星遍及全天球,其中亮于3等星的恒星占34.4%）,且弥补了部分天区恒星稀疏无法满足星图识别条件的不足,这将十分有利于提高星图识别速度和识别率;冗余导航星的剔除,为进一步提高星图识别率提供了可能。通过对恒星赋予特定的恒星标识,来确定该星参与哪些处理过程,如:双星、辅助导航星只参与星图识别过程;而参与姿态确定的则是非双星且位置精度阈值高的导航星,从而改善参考星的不精确引入的误差。（二）星图预处理为了尽可能的降低星图畸变及大气折射等因素的影响,采用质心法对实拍星图中心区域的星体进行质心提取,通过编程实现了星体质心计算结果的输出,供星图识别和姿态确定软件作为输入使用。为了估算姿态确定算法的精度,在Kenneth Daniel Diaz相机模拟器的基础上构建了适用于本研究的理想星相机模型,给出了理想星相机模拟器所用参数,详尽地图示了星相机模拟器所采用的坐标变换并推导了相应的姿态变换矩阵。通过坐标平移及转换把Hipparcos星表恒星从国际天球坐标系转换到像平面,进行了星图模拟,星图模拟的结果是模拟星图识别、姿态确定算法精度分析及影响星敏感器视轴姿态确定因素分析的基础。（三）改进的三角形星图识别算法采用改进的三角形星图识别算法,导航星库中识别特征存储的是两颗星之间星对角距,而不是导航三角形,节约了存储空间。全天球自主星图识别方法是星敏感器星图识别中最关键的技术,本研究在Christopher T.F. Kuehl的三角形星图识别的基础上进行了改进,即在一定角距容差和星对容差范围内和整个导航星角距表进行角距匹配,实时组成星三角,根据矢量法则剔除旋转方向不一致的星三角,把拥有公共顶点的星三角组合成星多边形。因为有伪星的存在,此时的最大多边形不一定是最佳匹配,本研究引入了投票算法的理念,对组成星三角的导航星进行匹配率统计,选取匹配率高的星多边形作为正确识别,从而排除了伪星的干扰。在局部天球星图识别方面,因为已知初始姿态,在一定角距容差和星对容差范围内和已知视轴附近的导航星角距表进行角距匹配,实时组成星三角,根据矢量法则剔除旋转方向不一致的星三角后,再把拥有公共顶点的星三角组合成星多边形。由于本研究过程中已经实现了姿态的快速反演,因此在组成星多边形之后根据识别结果进行了姿态反演,引入了已知姿态和反演姿态的比对环节,从而确保识别结果的正确性。通过蒙特-卡罗仿真对全天球星图识别和局部天球星图识别算法进行了仿真试验,结果表明全天球星图识别在有伪星存在的情况下星图识别率97.7%以上,同时识别视场内5颗星的识别速度在0.4秒以下。局部天球星图识别算法在识别速度和识别率上均高于全天球星图识别。对全天球星图识别算法进行了实拍星图试验,并给出了星图识别的正确性验证方法。星图识别试验过程中发现,星等容差和角距容差直接关系到星图识别的效率和速度,而其值的选择又和星敏感器的性能紧密相关。（四）姿态确定算法进行了改进的TRIAD算法和QUEST算法的探讨,这两种定姿算法各有千秋,在识别出的高精度导航星只有2颗的情况下,QUEST算法无法给出姿态,而改进的TRIAD算法可以进行三轴姿态的反演;但在识别出的高精度导航星多的条件下QUEST算法有明显的优势。对于实拍星图,基于导航星选取过程中赋予的恒星标识及星图识别结果给出了姿态确定算法的选取原则,即:根据导航星的恒星标识、识别出的恒星数目及不同定姿算法计算出的最小代价函数值的优劣进行最终姿态计算结果的选择输出。通过对1000幅随机视轴确定的模拟星图进行QUEST、改进的TRIAD、Kendiaz和QUEST₃Stars四种姿态算法的误差分析,发现QUEST算法的姿态计算结果最优、QUEST₃Stars其次、改进的TRIAD第三,Kendiaz算法误差最大。（五）影响姿态确定精度的若干因素分析通过星图模拟,对星点位置精度、参与姿态反演的恒星数目及恒星在视场中的分布对星敏感器视轴姿态计算精度的影响分析,得到如下新认识:对于特定视场大小和CCD尺寸而言,随着星点位置精度的提高,三轴姿态的精度在提高,之后达到临界值,此后无论如何提高星点位置精度,三轴姿态精度将无法提高。从统计学意义上来说,参与姿态反演的恒星数目越多,星敏感器视轴姿态计算的精度越高,但对于某一星图而言,需要具体问题具体分析,与定姿算法的选取有着密切的联系。恒星在视场中的分布对定姿算法至关重要,不同的定姿算法有不同的最佳（差）分布特征。改进的TRIAD算法的最佳恒星分布基本上需满足在视轴的两侧且和视轴成对称分布。QUEST算法的最佳星三角分布基本上需满足视轴位于星三角的重心附近;组成最佳星三角的恒星参与QUEST定姿的结果优于视场中的所有恒星参与QUEST定姿的结果;对于视场中恒星数目比较多的星图,有选择的挑选参与姿态计算的恒星将有助于提高姿态计算精度。（六）面向跨平台应用的研发（R&D）系统在以上研究的基础上,用C语言编制了一套面向跨平台应用的研发系统,该系统能直接用于真实星空观测的全天球自主星图识别及星敏感器视轴姿态确定,适用于微机和UNIX平台。该系统以质心确定结果为输入,进行星图识别后根据导航星的标识及预设的姿态优选算法进行星敏感器视轴姿态反演,最终输出星敏感器视轴的在赤道直角坐标系下的三轴姿态。通过对真实星空的星图识别来检验星图识别算法的正确性和可靠性。更多还原

【Abstract】 The new generation geostationary meteorological satellite FY-4 is a three-axis, body-stabilized satellite, on which the various remote instruments can be mounted. This highly improved the sounding efficiency and even allowed the primary sensors to“stare”at Earth. To achieve the high accuracy attitude, high precise star trackers will be used. The star trackers incessantly observe the portion of the sky and select the stars in view, then, send the stars to the Star-Identification Process. The process identifies the readings from the guide star catalog and the Attitude-Determination Process determines the exact attitude of the star tracker. According to the mounted attitude matrix of the star tracker, the attitude of the satellite can be determined. At Present, there is a large gap between the star tracker made in China and the international state-of-art star tracker on the aspects of precision, data updating rate and lifetime etc. This will bring a much higher challenge to the Attitude Control System （ACS） of the FY-4 satellite. According to the needs of the three-axis stabilized ACS of FY-4 satellite, this research has developed the Star-Identification algorithm and single-frame Attitude-Determination algorithm for the ground-based application system. Through Star-Identification and Attitude-Determination from the star map stores in the ground, the rate of attitude will be improved, the process of attitude change can be got, which will improve the accuracy of the satellite attitude and image navigation and registration.After the further study on star navigation technology of geostationary meteorological satellite, several algorithms of attitude determination based on star tracker for FY-4 meteorological satellite are researched and implemented. In the sub-catalog research, a new approach to choose guide star is proposed. In which the new concept of threshold of position precision and Additional Guide Star are introduced. In the star pattern recognition algorithm, the triangle-based Star-Identification algorithm , including entire celestial sphere autonomous Star-Identification algorithm and partial celestial sphere Star-Identification algorithm have been improved. In the attitude determination algorithm research aspect, under ideal conditions’ Optimized-TRIAD best stars distribution characteristic and ideal conditions’ QUEST best star triangle distribution characteristic is summarized. The major research and conclusions are as follows:（1） sub-catalog methodIn order to meet the requirement of high precision of attitude determination and image navigation of FY-4 meteorological satellite, set up a high precision guide star catalog for ground-base is very important. A method of chosen guide star is proposed. This method is based on the J1991.25 Hipparcos catalog and Pulkovo Compilation of Radial Velocities for Hipparcos Stars, through the rigorous treatment of epoch transformation, the star’s position of the FY-4 runtime epoch can be got. In order to got the more accurate Guide Star（with star flag 0）, the stars with magnitude brighter than 6.5 of J1991.25 Hipparcos catalog were chosen, after the rigorous epoch transformation, the J2000.0 Hipparcos sub-catalog was got, with the help of the information of Tycho-2 star catalog and SAO star catalog at the same epoch, the threshold of position precision is introduced, which can meet the precision of attitude determination in arc-second level. The Additional Guide Star （with star flag 5） is come into use for improving the accuracy and speed of star identification. After the cases analysis of unsuccessful star identification, giving the guide star which always cause false star identification a given star flag（such as 6）, which denote that the star is a Redundant Guide Star and can’t take part in the star identification processing. Through the introduction of Redundant Guide Star, the accuracy of star identification is improving. The distribution of Additional Guide Stars spread all over the whole sky, even the star sparse area. Moreover, in the 189 Additional Guide Stars, the proportion of star magnitude brighter than 3 reached to 34.4%, which is of great benefit to improve star identification. The attitude error will be introduced by the inaccuracy reference guide star, which is inevitable. The accuracy reference guide star, however, can improve the error of attitude determination.（2） Centroiding and Image SimulationAs far as possible to reduce the influences of the star pattern distortion and atmospheric refraction etc., the central regions of the star maps are centroiding, the positions of the stars in the star maps are output. It’s the input of the Star-Identification and attitude determination software. In order to estimate the accuracy of attitude determination algorithm, based on the camera simulator of Kenneth Daniel Diaz, an ideal star camera simulator suitable for this research is constructed. The parameters of the ideal star camera simulator are given, the coordinate transformations are well illustrated and the attitude transformation matrix is derived. Through the coordinate move and coordinate translation, the stars of Hipparcos star catalog were projected from the celestial sphere coordinate system to the image plane. The output of the star camera simulator is the basis of simulated star identification test, the attitude determination algorithm precision analysis and the factors analysis of which impact on star tracker boresight’s attitude determination.（3） Improved triangle Star identification algorithmFor the Improved triangle Star identification algorithm, the identify characteristics which stored in the guide star database are the separation angles of the two stars, rather than the guide star triangles, the storage space is saving. For the star tracker, the entire celestial sphere autonomous Star-Identification is the most essential technology. In this study, based on the Christopher T.F. Kuehl’s triangle Star-Identification, some improvement had made. With a certain angular and distance tolerance, the separation angle of the observed stars and that of the catalog stars are matched, then, the star triangles are composed in real time. According to the rules of vector rotation, the inconsistent rotation star triangles are removed. After that, put the star triangles with public vertex into star polygon. Because of the false star existence, the biggest polygon is not necessarily the optimum match, this study introduced the concept of voting algorithm to the stars of the matched star triangles, the star polygon who takes stars with high match rate as a right identification, thus removed the false star disturbance.In partial celestial sphere celestial Star-Identification, the initial attitude is known. With a certain angular and distance tolerance, the separation angle of the observed stars and that of the catalog stars are matched, then, the star triangles are composed in real time. According to the rules of vector rotation, the inconsistent rotation star triangles are removed. After that, put the star triangles with public vertex into star polygon. In this research, because of the attitude determination process had already realized, the identified stars of the polygon were send to the attitude determination process, the new attitude was got. After comparing the new attitude with the initial attitude, the accuracy of the Star-Identification is to be ensured.Through Monte-carol simulation, the entire celestial sphere autonomous Star-Identification algorithm and the partial celestial sphere Star-Identification algorithm are tested. For the former algorithm, it is shown that the star identification accuracy is higher than 97.7%, the speed of identify 5 stars in the same field of view is fast than 0.4 second. As far as the identification accuracy and the speed are concerned the latter is better than the former.（4）Attitude-Determination algorithmThe optimized TRIAD algorithm and QUEST algorithm are discussed, these two algorithms each has its good points. If there only 2 high accuracy guide stars are identified, the QUEST algorithm was unable to give the attitude, but the optimized TRIAD algorithm might. When the number of high accuracy guide stars identified is large, the QUEST algorithm has obvious superiority.According to night-sky observation, based on the guide star flag and the star- Identification information, the selection principle of attitude determination algorithm are given. That is: According to the guide star flag, the number of stars identified and the cost-function of two attitude determination algorithm to carry on the final attitude determination, the better attitude are chose as the output.Through Monte-carol simulation, the error of four cases of attitude determination algorithms is studied. The four cases of attitude determination algorithms namely QUEST, optimized TRIAD, Kendiaz and the QUEST₃Stars. It is shown that the QUEST algorithm’s attitude error is the most superior, the QUEST₃Stars next, optimized TRIAD third and the Kendiaz algorithm error is the biggest.（5）Attitude accuracy influencing factor analysisThrough the image simulation, several factors such as the accuracy of the stars’location ,the number of stars take part in attitude determination and the distribution of star in the field of view which influencing the attitude accuracy were studied. Some new findings are listed as follows:For the specific FOV size and the CCD size, with the enhancement of star Centroiding accuracy, the precision of three-axis attitude is enhanced. Afterward achieves the marginal value, hereafter with the enhancement of star Centroiding accuracy in any rate, the precision of three-axis attitude will not be improved.From statistics significance, the more stars take part in the attitude determination, the higher attitude precision of star tracker’s boresight is got. But speaking of certain star map, needs for concrete analysis, it depends on the attitude algorithm.The distribution of stars in the field of view is of great essential to the attitude determination algorithm, different algorithms have different best （/worst） distributed characteristic. Under ideal conditions, the optimized TRIAD algorithm’s best 2-stars distribution basically has the symmetrical distribution with the boresight and distribute on its two sides.Under ideal conditions, the QUEST algorithm’s best star triangle distribution basically satisfied the boresight locates nearby the gravity center of star triangle. When the number of stars in FOV is large, select stars satisfy the QUEST algorithm’s best distribution to take part in the attitude determination will help improve calculation accuracy.（6） Cross-platform application-oriented R & D systemOn the basis of the above provision, a set of cross platform research and development system with the C language has established. This system can use the real sky observation information to get the brighter star identified from the whole catalog and provides 3-axis pointing knowledge of the boresight relative to an inertial reference frame. It is suitable to PC and UNIX platforms. The night-sky Star maps are used for testing the correctness and reliability.更多还原