Cable joints are the location where partial discharges frequently occur. Aiming at the current insufficiency of partial discharge trend research and untimely early warning problems, a trend analysis and early warning method of partial discharge based on Mann-Kendall test method and long short-term memory (LSTM) neural network is proposed in this paper. Firstly, in order to clearly reveal the trend characteristics of the partial discharge volume, the Mann-Kendall test method is used to process the collected transient earth voltage (TEV) data, and quantitatively calculate the trend change and mutation point detection. Secondly, this paper proposes a comprehensive early warning model based on Mann-Kendall test method and LSTM algorithm. The model utilizes LSTM to predict TEV sequence amplitude, and Mann-Kendall is used to calculate trend parameters of predicted values, and the active early warning of partial discharge in cable joints is achieved through comprehensively considering TEV amplitude and trend parameters. The example results show that Mann-Kendall can clearly reveal the development trend of partial discharge, and the prediction effect of partial discharge data based on LSTM is good. The early warning model based on the two can better warn the partial discharge.