Total Organic Carbon (TOC) refers to the total amount of carbon contained in suspended or dissolved organic matter in water. It represents the concentration of organic matter in water by the mass of carbon contained in a unit volume of water. Total organic carbon can reflect more comprehensively the total amount of organic pollutants in the water. Monitoring total organic carbon can promote China to achieve the goals of “carbon peaking” and “carbon neutrality”, and it also has great significant meaning to the study of Chinas ocean earth carbon cycle. The national standard method for measuring water quality TOC mainly adopts the high-temperature catalytic oxidation method or wet oxidation method. Although the above two methods are accurate in measurement and have a high interpretability, they have disadvantages, such as complicated test methods, long measurement time, secondary pollution, and huge workforce and material resources waste. These methods can only be completed in the laboratory, so it is impossible to realize the in-situ online measurement of TOC. Therefore, it is -greatly significant for us to study the method of rapid and in-situ monitoring of TOC in water. This paper has established a single wavelength concentration detection model for TOC standard solution based on UV absorption spectra. Duo to more complex substance content of real water samples, ACO-PLS and SPA algorithms were used to select characteristic wavelengths and the performance of different spectral pretreatment methods, including S-G smoothing, min-max normalization, Standard Normal Variation (SNV), elimination of constant offset, derivative correction, were compared. The fast detection model of real water samples based on spectral absorption was established the least squares support vector machine algorithm (LSSVM) optimized by particle swarm optimization (PSO). The experimental results show that the modeling effect of SNV algorithm pretreatment is generally better than that of other pretreatment methods when a different numbers of characteristic wavelengths are selected. Moreover, the optimal number of characteristic wavelengths is generally 50 with different preprocessing algorithms because too many or too few wavelengths will reduce the modeling accuracy. The optimal modeling parameters are the SNV preprocessing method with 50 characteristic wavelength combinations selected by the ACO-PLS algorithm. The optimal PSO-LSSVM model result shows Rc=0.984 3, RMSEC=0.457 4, Rp=0.974 5, RMSEP=0.481 1. The optimal TOC detection was successfully applied to newly collected water, demonstrating the robustness of the model. ACO-PLS can effectively select the characteristic wavelength combination. Thus, the rapid determination of TOC in water quality based on UV-Vis absorption spectroscopy can be realized with the PSO-LSSVM algorithm, which provides a fast and pollution-free measurement scheme for TOC in water and provides theoretical support for the development of TOC sensors.