Scholarlink

<p>Faulty or inaccurate sensors can lead to increased energy consumption, disrupt control systems, and potentially cause damage to equipment. To tackle this challenge, we propose the Circle Mapping and Adaptive t-Distribution Runge–Kutta optimizer (CTRUN) to optimize the hyperparameters of a Long Short-Term Memory (LSTM) neural network designed explicitly for chiller sensor fault detection. CTRUN incorporates circle chaotic mapping for initialization and utilizes adaptive t-Distribution mutation. We compared CTRUN’s performance against several optimization algorithms, including the Runge–Kutta optimizer (RUN), particle swarm optimization (PSO), sparrow search algorithm (SSA), simulated annealing particle swarm optimization algorithm (SimuAPSO), adaptive spiral flying sparrow search algorithm (ASFSSA), secretary bird optimization algorithm (SBOA) and hippopotamus optimization (HO) algorithm. The results indicate that CTRUN significantly improves both convergence speed and optimization accuracy. Furthermore, we employ the Pearson correlation coefficient to assess feature correlations, allowing us to eliminate weakly correlated variables and boost diagnostic precision. The CTRUN-LSTM achieved an average improvement in fault detection rates of 13.89, 26.94, 33.17, and 12.05% for the tested sensors.</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

Chiller sensor fault diagnosis based on circle mapping and adaptive t-distribution Runge–Kutta algorithm optimized LSTM

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月6日

查看全文
  • Yuzhe Wang,
  • Jinglong Liu,
  • Feiyu Liu,
  • Xiaodong Duan

摘要

Faulty or inaccurate sensors can lead to increased energy consumption, disrupt control systems, and potentially cause damage to equipment. To tackle this challenge, we propose the Circle Mapping and Adaptive t-Distribution Runge–Kutta optimizer (CTRUN) to optimize the hyperparameters of a Long Short-Term Memory (LSTM) neural network designed explicitly for chiller sensor fault detection. CTRUN incorporates circle chaotic mapping for initialization and utilizes adaptive t-Distribution mutation. We compared CTRUN’s performance against several optimization algorithms, including the Runge–Kutta optimizer (RUN), particle swarm optimization (PSO), sparrow search algorithm (SSA), simulated annealing particle swarm optimization algorithm (SimuAPSO), adaptive spiral flying sparrow search algorithm (ASFSSA), secretary bird optimization algorithm (SBOA) and hippopotamus optimization (HO) algorithm. The results indicate that CTRUN significantly improves both convergence speed and optimization accuracy. Furthermore, we employ the Pearson correlation coefficient to assess feature correlations, allowing us to eliminate weakly correlated variables and boost diagnostic precision. The CTRUN-LSTM achieved an average improvement in fault detection rates of 13.89, 26.94, 33.17, and 12.05% for the tested sensors.