Although the smoothed particle hydrodynamics (SPH) method is effective for problems involving large deformation of object geometries, such as free surface flow, high computational efficiency is required to improve applicability to problems with a high number of particles. Reduced order modeling (ROM), a data-driven technique for accelerating numerical simulations, has been recently featured as a means to achieve more efficient computation. Viswanath et al.[1] developed ROM for simulation velocity field of Lagrangian free surface flow by using graph neural networks (GNN). However, further improvements are needed for engineering applications, such as pressure field seeking. This study aims to develop a deep learning-based ROM for simulating both velocity and pressure fields of free surface flow with high accuracy and reduced computational cost. ROM can be divided into an offline process of acquiring low-dimensional space and an online process of time evolution in low-dimensional space. In the offline process, Neural fields[2], which can obtain spatially continuous functions, were used to apply the model order reduction method to Lagrangian free surface flow data with spatially varying positions. The acquisition of low-dimensional fields is carried out in both velocity and pressure fields, making both subject to ROM. In the online process, the equation-driven method was used. Some previous studies have used GNN that require learning for time evolution, whereas Equation-driven methods do not require additional learning. Therefore, the learning phase is more efficient compared with previous research. Validation was carried out with the WaterDrop problem, which analyses the behaviour of falling water columns of different aspect ratios. Here, the speed-up of the proposed method is verified by comparing the proposed ROM with existing SPH methods. The results show that it is possible to speed up the process while maintaining a certain degree of accuracy.