UWB Indoor Localization Based on Artificial Rabbit Optimization Algorithm and BP Neural Network

Abstract

In the field of ultra-wideband (UWB) indoor localization, traditional backpropagation neural networks (BPNNs) are limited by their susceptibility to local minima, which restricts their ability to achieve global optimization. To overcome this challenge, this paper proposes a novel hybrid algorithm, termed ARO-BP, which integrates the Artificial Rabbit Optimization (ARO) algorithm with a BPNN. The ARO algorithm optimizes the initial weights and thresholds of the BPNN, enabling the model to escape local optima and converge to a global solution. Experiments were conducted in both line-of-sight (LOS) and non-line-of-sight (NLOS) environments using a four-base-station configuration. The results demonstrate that the ARO-BP algorithm significantly outperforms traditional BPNNs. In LOS conditions, the ARO-BP model achieves a localization error of 6.29 cm, representing a 49.48% reduction compared to the 12.45 cm error of the standard BPNN. In NLOS scenarios, the error is further reduced to 9.86 cm (a 46.96% improvement over the 18.59 cm error of the baseline model). Additionally, in dynamic motion scenarios, the trajectory predicted by ARO-BP closely aligns with the ground truth, demonstrating superior stability. These findings validate the robustness and precision of the proposed algorithm, highlighting its potential for real-world applications in complex indoor environments.

Keywords: BP neural network; UWB; artificial rabbit optimization algorithm; line-of-sight; non-line-of-sight.

Figure 1

Principle of the three-sided positioning algorithm.

Figure 2

Topology of BP neural network.

Figure 3

ARO-BP model of hybrid algorithm.

Figure 4

Schematic diagram of the LOS environment.

Figure 5

Actual coordinate position and UWB measurement data in the LOS environment: (a) 1.8 m and (b) 1.0 m.

Figure 6

Training dataset in the LOS environment: (a) 1.8 m and (b) 1.0 m.

Figure 7

Test dataset in the LOS environment: (a) 1.8 m and (b) 1.0 m.

Figure 8

Error curves between predictable and actual values in the LOS environment: (a) 1.8 m and (b) 1.0 m.

Figure 9

Histogram of the mean error of the test set in the LOS environment: (a) 1.8 m and (b) 1.0 m.

Figure 10

Schematic diagram of the NLOS environment.

Figure 11

Actual coordinate positions and UWB measurements in NLOS environment: (a) 0.45 m and (b) 0.15 m.

Figure 12

Training dataset in the NLOS environment: (a) 0.45 m and (b) 0.15 m.

Figure 13

Test dataset in the NLOS environment: (a) 0.45 m and (b) 0.15 m.

Figure 14

Error curves between predictable and actual values in the NLOS environment: (a) 0.45 m and (b) 0.15 m.

Figure 15

Histogram of the mean error of the test set in the NLOS environment: (a) 0.45 m and (b) 0.15 m.

Figure 16

Motion scenes.

Figure 17

Motion trajectory map in LOS environment.

Figure 18

Motion trajectory map in NLOS environment.