Due to the large number of hyperparameters, it is difficult to obtain the optimal system model for the bidirectional long short-term memory (Bi-LSTM) neural network. At the same time, considering the possibility of premature convergence in the Grey Wolf optimizer (GWO) algorithm, a single-lamp localization method using the GWO combined with particle swarm optimization (GWO-PSO) algorithm to optimize the Bi-LSTM neural network is proposed. By optimizing hyperparameters such as the learning rate and the number of hidden neurons in the network, the stability and positioning accuracy of the system are improved. Finally, the weighted K-nearest neighbors (WKNN) algorithm is used to optimize points with large errors to obtain more accurate positioning locations. The simulation results show that in an indoor environment of 3 mx3.6 mx3 m, the average positioning error of the proposed localization method is 3.57 cm, with 90% of the positioning errors within 6 cm.