Scholarlink

<p>In order to improve the accuracy and adaptability of indoor illuminance estimation in intelligent lighting systems, this paper proposes a CBKA-Transformer model based on an improved Black-Winged Kite Optimization Algorithm. In the proposed framework, CBKA improves hyperparameter optimization by combining chaotic population initialization, dynamic weight adjustment, and elite opposition-based learning, thereby enhancing global search ability and convergence efficiency. Meanwhile, time-aware positional encoding (TPE) embeds periodic temporal information, such as daily and seasonal illumination patterns, into the Transformer representation. The model further integrates multi-head self-attention and learnable spatial embedding to capture complex spatiotemporal dependencies among different monitoring points. The experiment was validated using multi season and multi weather illumination data collected through DIALux simulation. On the office environment test set, CBKA Transformer achieved an RMSE of 6.05&#xa0;lx and an R<sup>2</sup> of 0.978, with a prediction accuracy improvement of 34.5% compared to the standard Transformer. The ablation experiment showed that each module significantly improved performance, especially CBKA optimization reduced the number of convergence iterations of the model to 35 rounds. The results indicate that the method has good generalization ability and practical value.</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

Indoor illumination estimation based on improved black winged kite optimized transformer

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月19日

查看全文
  • Lirong Cheng

摘要

In order to improve the accuracy and adaptability of indoor illuminance estimation in intelligent lighting systems, this paper proposes a CBKA-Transformer model based on an improved Black-Winged Kite Optimization Algorithm. In the proposed framework, CBKA improves hyperparameter optimization by combining chaotic population initialization, dynamic weight adjustment, and elite opposition-based learning, thereby enhancing global search ability and convergence efficiency. Meanwhile, time-aware positional encoding (TPE) embeds periodic temporal information, such as daily and seasonal illumination patterns, into the Transformer representation. The model further integrates multi-head self-attention and learnable spatial embedding to capture complex spatiotemporal dependencies among different monitoring points. The experiment was validated using multi season and multi weather illumination data collected through DIALux simulation. On the office environment test set, CBKA Transformer achieved an RMSE of 6.05 lx and an R2 of 0.978, with a prediction accuracy improvement of 34.5% compared to the standard Transformer. The ablation experiment showed that each module significantly improved performance, especially CBKA optimization reduced the number of convergence iterations of the model to 35 rounds. The results indicate that the method has good generalization ability and practical value.