Aiming at the interference of the sampled ultra-weak Fiber Bragg Grating (FBG) reflectance spectrum, a high-speed peak seeking algorithm for large capacity ultra-weak sensor networks is proposed. In this algorithm, weight factors are introduced into the least squares fitting method to realize the Weighted Least Squares fitting (WLS) algorithm with optimized fitting coefficient of gaussian curve to locate the center wavelength. Then the central wavelength is modified by Asymmetric Gaussian correction (AG) to achieve high-precision peak location against noise interference. We propose WLS combined with AG (WLS-AG) algorithm. Through experiments, the peak error averages of the least square fitting algorithm, centroid algorithm, WLS algorithm and the WLS-AG algorithm proposed in this paper are compared and analyzed under different noises. The experimental results show that the WLS-AG algorithm works with high Signal-to-Noise Ratio （SNR）. The average error of 20 consecutive repetitive experiments is less than 1 pm, and the mean error is about 10 pm at low SNR. The detection error at different temperatures is within 1pm, which is the most stable and meets the requirements of ultra-weak FBG sensor system for precision demodulation.