The invention discloses a K mean value cluster-based optical fiber inertial measurement unit temperature model coefficient determination method. According to optical fiber inertial measurement unit temperature model coefficient first-order subsection cases, if the number of segments is determined, 1, according to a straight line determined through adjacent temperature points, a slope is calculated, 2, through a K mean value clustering algorithm, the slope is classified and a category code is determined, 3, according to the category code and the number of the segments, all segment point compositions are acquired, and 4, through a fitting residual error, the optimal segment point is determined, and if the number of segments is not determined, according to a sequence of 1 to the largest probable number of the segments, the optimal segment point under the condition of each one of the numbers of the segments is calculated through the above steps 1-4 until the residual error satisfies the requirements. Finally, according to the optimal segment points, the optical fiber inertial measurement unit temperature model coefficient is fitted. The method utilizes the K mean value clustering algorithm to automatically search the optimal segment point, overcomes random errors and repeated processes of manual subsection and effectively improves product temperature model coefficient calculating efficiency and reliability.