一种基于GACNN改进的室内可见光指纹定位算法

An indoor visible light fingerprint localization algorithm based on GACNN’s improvement

为提高室内可见光定位系统性能,提出了基于遗传算法训练卷积神经网络(Genetic Algorithm Convolutional Neural Network,GACNN)的室内可见光指纹定位算法。该算法引入一维卷积神经网络学习模型,针对卷积神经网络的超参数设置,利用遗传算法对卷积神经网络进行训练,将超参数进行二进制编码后采用精英遗传算法对CNN进行训练,来解决卷积神经网络模型参数调节依靠经验和模糊最优化的过程。实验结果表明:在室内4 m×4 m×2.5 m的定位场景下,定位算法可以获得平均定位误差4.11 cm的定位精度。相较于卷积神经网络定位算法,平均定位误差降低了25%。对比分析了不同室内可见光定位算法的性能,验证了算法的技术优势。

In order to improve the performance of indoor visible light localization system,an indoor visible light fingerprint localization method based on Genetic Algorithm-Convolutional Neural Network(GACNN)training is proposed.The algorithm introduces a one-dimensional convolutional neural network learning model,uses the genetic algorithm to train the convolutional neural network according to the hyperparameter setting of the convolutional neural network,binars the hyperparameters and then uses the elite genetic algorithm to train the CNN to solve the process of relying on experience and fuzzy optimization of the parameter adjustment of the convolutional neural network model.The experiment results show that in the indoor positioning scenario of 4 m×4 m×2.5 m,the positioning accuracy of the average positioning error of 4.1 cm can be obtained by using the proposed positioning algorithm.Through simulation experiments,compared with the convolutional neural network positioning algorithm,the average positioning error is reduced by 25%.The performance of different indoor visible light positioning algorithms is compared and analyzed,and the technical advantages of this algorithm are verified.