This paper reports research based on pneumatic subsoiling and the design of a pneumatic subsoiling mechanism to overcome the problems of high resistance and high energy consumption of subsoiling. By analysing soil‐specific resistance, soil disturbance rate and soil bulkiness under different air pressure conditions, it is concluded that pneumatic subsoiling can effectively break the soil plough pan and reduce resistance to subsoiling. In order to analyse the impact of air pressure on subsoiling, in this study, principal component analysis was used to analyse the pneumatic subsoiling disturbance parameters (working air pressure, working depth and working speed), and the test results show that the contribution of air pressure to subsoiling resistance and subsoiling disturbance surface reached 24% and 25%, respectively. An orthogonal test was used to analyse the specific resistance of subsoiling, and its significance coefficient is 0.95. Long short‐term memory neural networks (LSTM) and bidirectional long short‐term memory neural networks Bi‐LSTM. are used to predict the cracks on the disturbed surface of subsoiling. LSTM is a method to predict future occurrence using time series data, which can be used to predict the cracks on the disturbed surface of soil, while Bi‐LSTM network is an innovative computing paradigm, which learns bidirectional long‐term correlation between time step and sequence data, to predict the trend of fissures on the disturbed soil surface. The RSME of LSTM and Bi‐LSTM are 4.80 and 6.55, and their determinative factor R2 is 0.95 and 0.94 respectively, which indicates that LSTM and Bi‐LSTM can effectively predict the cracks of pneumatic subsoiling. By analysing the specific resistance of pneumatic subsoiling, it can be shown that pneumatic subsoiling can reduce subsoiling resistance and expand the disturbance surface of subsoiling so as to achieve the effects of subsoiling, drag reduction and reduction of fuel consumption.