Accurate detection of boric acid concentration is a crucial factor in ensuring the safety of nuclear power plants. Raman spectroscopy can achieve online, non-destructive, and rapid detection of boric acid concentration, offering significant application value. However, the baseline signal in Raman spectroscopy is susceptible to strong interference from laser power fluctuations, changes in the position of the Raman probe, and variations in sample temperature. To address the issue of insufficient stability in traditional Raman spectral baseline correction algorithms when handling complex baselines, we proposed a Genetic Algorithm-based Adaptive Smoothed Least Squares Baseline Correction Algorithm (GA-ASLS). This algorithm maintains high stability when processing complex baseline Raman spectral data, ensuring a good signal-to-noise ratio. Although it sacrifices a small amount of quantitative accuracy, it significantly improves generalization capability. The stability of the algorithm was evaluated by calculating the standard deviation and relative standard deviation of the characteristic peaks in the boric acid spectrum after baseline removal. The results show that the standard deviations and relative standard deviations for the PFBC model, QSBC model, and GA-ASLS model after baseline removal were 39.80 (1.22%), 35.96 (1.07%), and 27.34 (0.68%), respectively. These results indicate that the stability of the GA-ASLS algorithm is significantly better than that of the PFBC and QSBC models. Quantitative analysis of the baseline-corrected spectra using Multiple Linear Regression (MLR) on the GA-ASLS model demonstrated that the prediction error was less than 25 ppm for a boric acid concentration of 1000 ppm, and less than 2.5% for concentrations ranging from 1000 to 4500 ppm. This further confirms the high precision and reliability of the GA-ASLS algorithm in practical applications.