Simulating human mobility realistically and generating large-scale, high-quality trajectories are crucial for various location-based applications such as traffic management, epidemic spreading analysis, and location privacy protection. While the most popular model-free methods succeed by directly learning distribution of real-world data, they struggle to produce high-quality mobility data without leveraging the domain knowledge of human mobility. Moreover, such model-free methods primarily rely on auto-regressive paradigms, usually accompanied by error accumulation problem. To address the issues, we propose a model-free Domain-Knowledge Enhanced Generative Adversarial Network (DKE-GAN), which efficiently combines domain knowledge of urban context with model-free learning paradigm to generate high-quality mobility data. In addition, we incorporate reinforcement learning into the training process, thereby effectively alleviating the error accumulation. Furthermore, we introduce Trajectory Representation Learning (TRL) to convert noise-carrying raw trajectories into low-dimensional representation vectors for fully mining human mobility patterns. Extensive experiments conducted on two representative real-world mobility datasets demonstrate that our proposed method outperforms six state-of-the-art baselines, significantly achieving performance improvements in simulating human mobility.

In China, a majority of the proven crude oil reserves are found in clastic rock reservoirs, which typically exhibit low natural energy levels. Water injection has become the most widely adopted technique for maintaining reservoir pressure and enhancing oil recovery in such formations. However, conventional water injection strategies heavily rely on empirical knowledge, often failing to accurately characterize the dynamic inter-well connectivity between injection and production wells. This limitation hinders the effective management of fluid injection and production processes. To address this challenge, we propose an intelligent optimization method for water allocation in high-water cut, low-permeability reservoirs. Our approach employs a Bidirectional Long Short-Term Memory (BiLSTM) neural network to learn the complex patterns from historical injection data in a data-driven manner. Furthermore, we design a well distance and time joint attention mechanism, which is integrated after the dual BiLSTM layers to enhance the model’s ability to capture the critical dynamic relationships among wells. This mechanism decouples temporal pattern recognition and the spatial physical constraints, laying the foundation for interpretable injection strategy optimization. We name this architecture “AttBiLSTM”, which is designed for optimizing injection strategies for individual layers in separate-layer water injection wells (The layer refers to the basic geological unit or flow unit within a vertically heterogeneous reservoir that is delineated and requires independent water injection regulation). Using field data from the Xinjiang Oilfield, we validate the proposed method and compare its performance against traditional water injection schemes and mainstream data-driven models. The experimental results demonstrate that the AttBiLSTM model effectively establishes a nonlinear mapping between the injection volumes and oil production rates, showing strong performance in both production prediction and injection optimization. An independent numerical reservoir simulation verification confirms that the optimized scheme increases well group oil production by over 3.6%, with no premature water breakthrough risk in a 5-year development cycle. This study provides a novel and practical technical framework for efficiently developing low-porosity, low-permeability, and highly heterogeneous reservoirs.