Chaotic time sequence prediction method based on attention mechanism deep learning

Abstract

The invention belongs to the technical field of chaotic systems. The invention discloses a chaotic time sequence prediction method based on attention mechanism deep learning. The method comprises thefollowing steps: (1) constructing a chaotic time sequence data set; (2) carrying out phase space reconstruction on a chaotic time sequence; (3) training chaotic time sequence data by using an LSTM neural network model; (4) constructing a prediction-based attention mechanism model, (5) constructing an offline training model, and (6) carrying out online prediction. The chaotic time sequence prediction method based on attention mechanism deep learning is clear in model structure, has a reference value, and can be applied to the aspects of financial market prediction or energy prediction and the like of a chaotic system.

Description

technical field [0001] The invention relates to a chaotic time series prediction method based on attention mechanism deep learning, belonging to the technical field of chaotic systems. Background technique [0002] Chaotic system is a kind of orderly high-dimensional complex nonlinear dynamical system generated from disordered motion. The discrete situation of chaos is often manifested as chaotic time series. Chaotic time series is a time series with chaotic characteristics generated by chaotic systems. Chaotic time series contains very rich dynamic information of the system. Chaotic time series is the path from chaos theory to reality. A bridge to the world is an important research field of chaotic systems. According to Takens phase space delay reconstruction theorem, the internal law of chaotic system can be reconstructed and predicted by chaotic time series. How to choose or construct a forecasting model is one of the key issues in chaotic time series forecasting. [0...

AI Technical Summary

Problems solved by technology

Nevertheless, the traditional neural network learning algorithm based on gradient descent still has some obvious drawbacks and limitations in the application of chaotic time series forecasting, including: (1) These models converge slowly and tend to fall into local optimum, which affects the Predictive Accuracy and Wide Application

(2) Due to the static network, these models have limitations in identifying chaotic dynamical systems

(3) For these networks, determining an appropriate number of hidden nodes is also a challenge

Although many chaotic time series prediction methods based on deep learning have emerged, especially the long-short-term memory neural network LSTM, although LSTM can capture long-term dependencies and make up for the problems of gradient disappearance and gradient explosion in the training process of recurrent neural network RNN, but Its attention to the sliding window step size of each hidden layer is consistent, which may cause distraction problems and affect the prediction effect of chaotic time series

Images