Research on Long-Term Time Series Daily Maximum Temperature Prediction Based on a Conv1D-LSTM Hybrid Model
This paper addresses the challenges traditional methods face in processing high-dimensional data and capturing the nonlinear patterns and complex dynamic features in temperature data. A hybrid model based on Convolutional Neural Networks (Conv1D) and Long Short-Term Memory networks (LSTM) is proposed for long-term high-temperature prediction. The dataset consists of meteorological data from Beijing spanning 2014 to 2023,including features such as weather,daily minimum and maximum temperatures,and wind direction. Through feature engineering,weather and wind direction features were encoded,and temperature features were normalized. The proposed Conv1D-LSTM hybrid model innovatively integrates Conv1D to capture local features in the time series and integrates LSTM to learn long-term dependencies. Compared with traditional models,the hybrid model's Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) decreased by approximately 17.3％ and 20.5％,respectively,while the R2 score increased by approximately 1.06％,demonstrating higher prediction accuracy and generalization capability.
daily maximum temperature predictionConv1D-LSTM hybrid modellong-term time seriesprediction accuracy
万方数据
