Optimization of LED Luminaire Life Prediction Algorithm by Integrating Feature Engineering and Deep Learning Models

With the wide application of LED luminaires in various fields, it has become particularly important to accurately predict their lifetime. The lifetimes of LED luminaires are affected by a variety of factors, including temperature, current, voltage, light intensity, and operating time, and there are complex interactions among these factors. Traditional prediction methods are often difficult to capture these nonlinear relationships, so a more powerful prediction model is needed. In this study, we aim to develop an efficient life prediction model for LED luminaires, and propose a hybrid neural network structure that incorporates a convolutional neural network (CNN), a long short-term memory network (LSTM), and an attention mechanism by combining feature engineering and deep learning techniques. In the research process, we first collected the operation record data provided by a well-known LED lighting manufacturer and performed detailed data preprocessing, including missing value processing, outlier detection, normalization/standardization, data smoothing, and time series segmentation. Then, we designed and implemented several benchmark models (e.g., linear regression, support vector machine regression, random forest regression, and deep learning model using only LSTM) as well as the proposed hybrid neural network model. Through a detailed experimental design including parameter setting, training and testing, we evaluate the performance of these models and analyze the results. The experimental results show that the proposed hybrid neural network model significantly outperforms the conventional model in key performance metrics such as root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R²). In particular, the hybrid model outperforms in terms of Mean Absolute Percentage Error (MAPE) and Maximum Absolute Error (Max AE). In addition, through cross-validation and testing on different datasets, the model shows stable performance under various environments and conditions, verifying its good generalization ability and robustness.