改进型神经网络在室内三维定位中的应用研究 *

针对室内复杂环境下无线信号反射，折射，多径效应，噪声等干扰，传统的对数距离路径损耗模型无法精确求出信号接收距离d的问题进行研究，提出一种改进的多策略粒子群神经网络模型，该网络采用反向学习策略，混沌惯性权重策略以及扰动策略对神经网络的拓扑结构和连接权值阈值进行优化，从而捕获Wi-Fi信号接收强度RSSI与接收距离d之间的非线性关系，然后结合修正的三维加权交点质心算法求解出未知节点坐标。实验结果表明，对比BP神经网络和遗传神经网络，该网络对RSSI-d的拟合曲线更光滑，拟合结果更加接近于真实值，对比定位精度分别提高了79%和73%，网络定位误差更低，抗干扰能力更强。

Research of indoor 3D localization based on improved neural network

Under the complicated environment interference of indoor wireless signal such as reflection, refraction, multipath effect, noise and so on, the traditional logarithm distance path loss model cannot accurately calculate the signal receiving distance, this paper proposed an improved multi strategy particle swarm neural network model, the network used the reverse learning strategy, chaotic inertia weight strategy and disturbance strategy to optimize topological structure, weights and thresholds in order to capture the nonlinear relationship between Wi-Fi received signal strength RSSI and the receiving distance d, and then combined with the modified weighted centroid algorithm to calculate 3D intersection point coordinates of unknown nodes. Compared with the BP neural network and genetic neural network, the experimental results show that the network of RSSI-d curve is more smooth, fitting results closer to the true value, the positioning accuracy are improved by 79% and 73% respectively. At the same time, the network position has lower error and stronger anti-interference ability.