基于自组织模糊神经网络的大功率LED调光模型
2021年电子技术应用第12期
李纪宾,饶欢乐,王 晨,钱依凡,洪哲扬
杭州电子科技大学 自动化学院,浙江 杭州310018
摘要:大功率LED光度输出不仅与操作电流大小有关,且受传热过程的时滞时变不确定因素影响难以预测。针对传统机理建模存在参数提取困难、模型适应性弱等缺点,提出基于模糊神经网络建模算法,从而构建以操作电流、热沉温度、环境温度为输入,光通量为输出的调光模型。模型结构和参数依据在线数据进行调整,通过递推学习,模糊规则得到增量式完善,进而不断逼近实际动态过程。结果表明,利用该方法构建的调光模型与参考模型理论值相对误差小于3%,与其他模型相比,结构更加紧凑,预测精度更高。
中图分类号:TN364+.2
文献标识码:A
DOI:10.16157/j.issn.0258-7998.201125
中文引用格式:李纪宾,饶欢乐,王晨,等. 基于自组织模糊神经网络的大功率LED调光模型[J].电子技术应用,2021,47(12):105-109.
英文引用格式:Li Jibin,Rao Huanle,Wang Chen,et al. Dimming model of high-power LED based on self-organizing fuzzy neural network[J]. Application of Electronic Technique,2021,47(12):105-109.
文献标识码:A
DOI:10.16157/j.issn.0258-7998.201125
中文引用格式:李纪宾,饶欢乐,王晨,等. 基于自组织模糊神经网络的大功率LED调光模型[J].电子技术应用,2021,47(12):105-109.
英文引用格式:Li Jibin,Rao Huanle,Wang Chen,et al. Dimming model of high-power LED based on self-organizing fuzzy neural network[J]. Application of Electronic Technique,2021,47(12):105-109.
Dimming model of high-power LED based on self-organizing fuzzy neural network
Li Jibin,Rao Huanle,Wang Chen,Qian Yifan,Hong Zheyang
School of Automation,Hangzhou Dianzi University,Hangzhou 310018,China
Abstract:The luminosity output of high-power LED system is not only related to the current, but also hard to be predicted due to the uncertain nonlinear characters of thermal process. In view of the difficulties in extracting the parameters of the mechanism model and poor adaptability, an online modeling method was proposed to construct a fuzzy neural network with ambient temperature, heat sink temperature and operating current as input,and luminous flux as output. The model structure is self-organized and adjusted according to clustering analysis and error evaluation criteria. EKF algorithm and recursive least square method are used to learn network parameters. Through recursive learning, the rule is improved incrementally so that the model can approximate the actual system process as fast as possible. Validity of the algorithm is verified in a typical nonlinear system. Results show that the relative error between the theoretical values of the photometric prediction model and the reference model is less than 3%. Comparing with other model, this model has more compact structure and better generalization performance.
Key words :high-power LED;PET model;self-organizing fuzzy neural network;structure identification;parameter learning
0 引言
相较于传统光源,大功率LED具有高光效和灵活可控等优势,在提供交互式或动态照明方面颇具潜力,如建筑照明[1]、太阳光模拟器[2]等。这类光源通常要求光度输出宽范围动态可调,并且快速达到预定的精度要求。尽管LED自身开关特性可达兆赫兹,但由于系统散热存在时滞、时变不确定特性,使得光度输出规律难以预测。构建可分析、可计算和执行的调光模型对实现更加精细化的调光控制具有重要意义。
经典光电热[3]理论表明LED结温、光通量、电流存在多参数耦合关系。而后,Tao[4]等人通过机理分析,构建动态光电热模型,用于计算光通量输出随系统温升的衰减变化。文献[5]~[6]考虑环境温度的热因素影响,构建不同操作功率下的线性扰动模型,设计了温度前馈补偿器,以保证光度的恒定输出。文献[7]建立了基于状态空间表达的线性预测模型,便于移植到低成本控制器中去。文献[8]采用多项式插值方法辨识不同驱动电流下的传递函数的零极点增益,构建了线性参数时变模型,但该方法需预先设置整个工作范围的操作条件,计算量较大。尽管LED物理机制明确,但多数模型[3-6]基于等效阻容网络分析,部分物理量(如结温)并不易于测量,且模型采用离线设计,在长时运行或环境变化较大的条件下将存在失配问题。
本文详细内容请下载:http://www.chinaaet.com/resource/share/2000003881。
作者信息:
李纪宾,饶欢乐,王 晨,钱依凡,洪哲扬
(杭州电子科技大学 自动化学院,浙江 杭州310018)
此内容为AET网站原创,未经授权禁止转载。