Author: Zen and the Art of Computer Programming
1 Introduction
Regarding time series data forecasting and regression, it is very important in all walks of life. With the continuous breakthrough of artificial intelligence, machine learning, deep learning and other technologies, the methods of forecasting and regression of time series data have also received extensive attention. In recent years, deep learning technology has become an effective tool for solving complex tasks, especially in the field of time series prediction and regression. Many excellent models have been proposed, such as recurrent neural network (RNN), long short-term memory network (LSTM), transformer network ( Transformer), etc., and achieved remarkable results. This article will introduce several popular time series forecasting and regression methods and related latest papers.
Time series forecasting and regression is the use of historical data to speculate on future data, which involves analysis in the time dimension. It mainly includes single-step, multi-step, time series and structured prediction. Single-step forecasting refers to using the value of one or more variables in the past period of time to predict the value of a variable at the next time point; multi-step forecasting refers to using the value of one or more variables in a certain period of time in the past to predict the next time The value of one or more variables at multiple time points down; time series forecasting is to predict the value of one or more variables at multiple time points in the future from the entire time series; structured forecasting is based on the input data. Structural information to make predictions, such as collaborative filtering, factorization, etc.
Time series data has the characteristics of diversity, discontinuity, and variable length. Therefore, traditional time series analysis methods cannot effectively deal with this type of data, and it is necessary to rely on deep learning technology to achieve related tasks.
In this article, we will first introduce the basic knowledge of time series data forecasting and regression, and then list different models for different tasks, and elaborate on their characteristics, advantages and disadvantages, and application scenarios. Finally, the specific implementation process and formula derivation of these models will be given, as well as how to use them in practical tasks.
2. Related background
What is the time series data that needs to be clarified first? The definition of time series data refers to the pattern of occurrence of a process or event over time, that is, as time goes by, the observer