To improve the measurement accuracy and measurement stability of the distributed fiber Raman temperature measurement system, a temperature demodulation method that self-compensates for fiber loss, fiber dispersion, and system fluctuation was presented and experimentally verified. The loss coefficient between the Stokes and Anti-Stokes backscatter signals was calculated in real time by sampling value, and the difference between them was dynamically corrected. This prevents the temperature accuracy from decreasing as the fiber length increases. A linear interpolation algorithm was used to correct the misalignment of the Stokes and anti-Stokes backscatter signals due to the dispersion effect of the optical fiber, eliminating temperature anomaly points caused by dispersion on both sides of the heating point, thus improving temperature measurement accuracy. Finally, the fluctuation of the system was corrected according to the sampling value and the environmental temperature value. Thus, the fluctuation range of the temperature demodulation value was reduced. Experimental results indicated that the error range of temperature measurement decreased from -2.72-7.24% to -0.34-1.04% when the fiber distance was 8 km. Temperature fluctuations were reduced from -0.0128-1.6181 ℃ to -0.8991-0.6476 ℃ across the fiber.