Multi-Feature Temporal Prediction Based on Hybrid LSTM Models

Multi-feature time series prediction is in a more important position in the field of time series prediction, which provides richer feature representations for the predicted variables by integrating the characteristics of different features, making it widely used in practical applications. In this paper, a hybrid LSTM model is proposed by combining the popular LSTM algorithms, using XGBOOST to extract the key features first, then combining CNN with BILSTM and introducing the weighted attention mechanism to obtain more critical information, and finally applying the ARIMA model to process the residuals to get the final predicted values. The proposed LSTM hybrid model has a better fitting effect compared with other common models, can better predict according to the characteristics of the remaining features, and can be widely used in multi-feature time-series prediction.