A Prediction Method Based on LED Chromaticity Shift

The chromaticity shift is the primary index for assessing the reliability of LED products. However, existing prediction models for chromaticity shift suffer from insufficient accuracy and prolonged data collection times. This paper presents a novel method based on self-attentive nested Long Short Term Memory (LSTM) for estimating LED chromaticity offset to address these challenges. Leveraging the memory structure and self-attentive mechanism of the nested LSTM, this approach effectively captures the features of LED chromaticity offset data and their dependencies, while mitigating issues such as redundancy or loss of output information inherent in nested LSTMs. Experimental data on LED chromaticity shift is collected according to the measurement conditions specified in the IES LM-80-15 standard. Results demonstrate that the proposed method outperforms existing prediction techniques in terms of accuracy and significantly reduces testing time for LED products.