The Yangtze economic belt accounts for more than 40% of China's total populaton and economy, and the urban agglomeration in the middle reaches of the Yangtze River is more vulnerable to the threat of flood disasters due to the comprehensive influence of geographical location and hydrological conditions. The water level of the Lianhuatang station is served as the index control station of the flood storage and detention area in the middle reaches of the Yangtze River. When the water level of the Three Gorges Reservoir reaches 155.0 m, if the water level of the Lianhuatang station reaches 34.40 m and continues to rise, flood diversion measures should be taken in the flood storage and detention area near Chenglingji. While under the dual influence of the river form and the relationship between the river and the lake (the Yangtze River and Lake Dongting), the water level of the Lianhuatang station presents tidal irregular periodic fluctuation (commonly known as "false tide"), which brings decision-making difficulties to the flood storage and detention area. To study the related factors of flood level fluctuation phenomenon and water level filtering method in the Chenglingji reach of the Yangtze River at the outlet of Lake Dongting, this paper analyzed the self-recorded water level data of Lianhuatang water station in the last 5 years and 18 months (June-August 2016, July-October 2017, May-August 2018, June-August 2019, and June-September 2020). Firstly, the single-day water level fluctuation value D>0.05 m is defined as the criterion for the occurrence of "false tide". And the Binary Logistic regression analysis showed that during the period of high water level (Z_j ≥ 34.00 m) at Jianli station, when the water level drop from Jianli to Lianhuatang (ΔZ_jl) is greater than 2.75 m, the proportion of "false tide" is 100%. The comprehensive correction value of model prediction results was 96.4%, indicating that ΔZ_jl was the main factor influencing the false tide. Then, two data filtering methods, fast fourier transform and local regression, are used to process the water level fluctuation sequence, which shows that both of the two methods are applicable to the analysis of wave shape data with different amplitudes, and are not affected by short-term data loss and instantaneous water level value, which may have great application value. From the perspective of the fluctuation of the error value, the locally weighted regression algorithm has a smaller fluctuation of the error value and is more suitable for the processing of the fluctuating water level data.