Scholarlink

<p>The performance of long short-term memory (LSTM) networks depends mainly on the hyper-parameters, which significantly affects the training stability, convergence, and generalization. This study focuses on the feasibility of using LSTM network to predict the thermally induced error in turning center, resulting in spindle drift. The predictability of the model is improved, by optimizing the hyper-parameters in LSTM using grey wolf optimizer (GWO). The hyper-parameters considered include hidden layers, neurons per layer, batch size, number of epochs, learning rate, and dropout rate. The optimization is aimed at minimizing the mean squared error (MSE) of the predicted thermal error. The prediction model developed using the GWO optimized LSTM algorithm is trained and tested using the data from the turning center. GWO-optimized LSTM model provided a more accurate prediction than the non-optimized LSTM model upon comparison with the actual cutting data. MSE has reduced from 152 to 6, upon using GWO-optimized LSTM model compared to non-optimized LSTM model. The robustness of GWO is compared with other metaheuristic algorithms like particle swarm optimization (PSO) and genetic algorithm (GA). GWO-LSTM performed better in convergence with least MSE (i.e) 5.27 compared to other algorithms providing 10 and 12 respectively.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Improved thermal error prediction in turning center by optimizing the hyper-parameters in LSTM model using GWO

  • 期刊论文

  • 发表日期：

    2026年5月7日

查看全文
  • M. Imran Alam,
  • N. Rino Nelson

摘要

The performance of long short-term memory (LSTM) networks depends mainly on the hyper-parameters, which significantly affects the training stability, convergence, and generalization. This study focuses on the feasibility of using LSTM network to predict the thermally induced error in turning center, resulting in spindle drift. The predictability of the model is improved, by optimizing the hyper-parameters in LSTM using grey wolf optimizer (GWO). The hyper-parameters considered include hidden layers, neurons per layer, batch size, number of epochs, learning rate, and dropout rate. The optimization is aimed at minimizing the mean squared error (MSE) of the predicted thermal error. The prediction model developed using the GWO optimized LSTM algorithm is trained and tested using the data from the turning center. GWO-optimized LSTM model provided a more accurate prediction than the non-optimized LSTM model upon comparison with the actual cutting data. MSE has reduced from 152 to 6, upon using GWO-optimized LSTM model compared to non-optimized LSTM model. The robustness of GWO is compared with other metaheuristic algorithms like particle swarm optimization (PSO) and genetic algorithm (GA). GWO-LSTM performed better in convergence with least MSE (i.e) 5.27 compared to other algorithms providing 10 and 12 respectively.