Abstract: Abstracts: In oil and gas drilling engineering, the dynamic optimization of drilling parameters is a core process for improving drilling efficiency and reducing operational costs. However, existing methods for full-well-section parameter optimization all adopt a fixed "continuous tracking throughout the entire process" mode, lacking the adaptability for flexible start-stop operation. Consequently, they cannot meet the practical requirement of "on-demand optimization and real-time solution" in field operations. To address this issue, this paper proposes a memoryless dynamic parameter optimization method for the full well section based on a neural network architecture. This method overcomes the convergence difficulty of conventional recurrent neural networks (RNNs) and constructs a high-precision mapping model between the temporal data flow of drilling parameters and the rate of penetration (ROP) using a long short-term memory (LSTM) network. ROP is then set as the optimization objective. By constructing a multi-dimensional time-series dataset of drilling parameters, the method enables real-time parameter optimization for any well section without inheriting the state information of previous sections (i.e., on-demand start-stop and independent solution, independent of historical optimization results). Experimental results show that, in complex formation environments, the parameter optimization accuracy of the proposed method is about 90% higher than that of traditional RNN models, and it offers greater engineering convenience and practicality compared with the conventional "full-process continuous tracking" mode, providing reliable technical support for the engineering application of intelligent drilling systems.