The Time Series Forecasting Problem
There are many different types of tasks on time series data:
 classification,
 anomaly detection,
 forecasting.
Forecasting Problem
A time series forecasting problem can be formulated as the following.
Given a dataset $\mathcal D$, with
 $y^{(i)}_t$, the sequential variable to be forecasted,
 $x^{(i)}_t$, exogenous data for the time series data,
 $u^{(i)}_t$, some features that can be obtained or planned in advance,
where ${}^{(i)}$ indicates the $i$th variable, ${}_ t$ denotes time. In a forecasting task, we use $y^{(i)} _ {tK:t}$, $x^{(i) _ {tK:t}}$, and $u^{(i)} _ {tK:t+H}$, to forecast the future $y^{(i)} _ {t+1:t+H}$.
A model $f$ will use $x^{(i)} _ {tK:t}$ and $u^{(i)} _ {tK:t+H}$ to forecast $y^{(i)} _ {t+1:t+H}$.
The above formulation is mostly focusing on the point forecasts. Alternative, the [[probabilistic view]] Prediction Space in Forecasting In a forecasting problem, we have $\mathcal P$, the priors, e.g., price and demand is negatively correlated, $\mathcal D$, available dataset, $Y$, the observations, and $F$, the forecasts. Information Set $\mathcal A$ The priors $\mathcal D$ and the available data $\mathcal P$ can be summarized together as the information set $\mathcal A$. Under a probabilistic view, a forecaster will find out or approximate a CDF $\mathcal F$ such that1 $$ \mathcal F(Y\vert \mathcal D, \mathcal P) \to F. … of a forecasting problem has been a very hot topic in recent years.
The Time Delay Embedding Representation
The time delay embedding representation of a time series forecasting problem is a concise representation of the forecasting problem ^{1}.
For simplicity, we only write down the representation for a problem with time series $y_{1}, \cdots, y_{t}$, and forecasting $y_{t+1}$. We rewrite the series into a matrix, in an autoregressive way,
$$ \begin{align} \mathbf Y = \begin{bmatrix} y_1 & y_2 & \cdots & y_p &\Big & {\color{red}y_{p+1}} \\ y_{1+1} & y_{1+2} & \cdots & y_{1+p} &\Big & {\color{red}y_{1+p+1}} \\ \vdots & \vdots & \ddots & \vdots &\Big & {\color{red}\vdots} \\ y_{ip+1} & y_{ip+2} & \cdots & y_{i} &\Big & {\color{red}y_{i+1}} \\ \vdots & \vdots & \ddots & \vdots &\Big & {\color{red}\vdots} \\ y_{tp+1} & y_{tp+2} & \cdots & y_{t} &\Big & {\color{red}y_{t+1}} \\ \end{bmatrix} \end{align} $$
which indicates that we will use everything on the left, a matrix of shape $(tp+1,p)$, to predict the vector on the right (in red).
Methods of Forecasting Methods
T. Januschowsk et al proposed a framework to classify the different forecasting methods.^{2}

Hewamalage2022 Hewamalage H, Ackermann K, Bergmeir C. Forecast Evaluation for Data Scientists: Common Pitfalls and Best Practices. arXiv [cs.LG]. 2022. Available: http://arxiv.org/abs/2203.10716 ↩︎

Januschowski2019 Januschowski T, Gasthaus J, Wang Y, Salinas D, Flunkert V, BohlkeSchneider M, et al. Criteria for classifying forecasting methods. Int J Forecast. 2020;36: 167–177. doi:10.1016/j.ijforecast.2019.05.008 ↩︎
 Lim B, Zohren S. Time Series Forecasting With Deep Learning: A Survey. arXiv [stat.ML]. 2020. Available: http://arxiv.org/abs/2004.13408
 Hewamalage2022 Hewamalage H, Ackermann K, Bergmeir C. Forecast Evaluation for Data Scientists: Common Pitfalls and Best Practices. arXiv [cs.LG]. 2022. Available: http://arxiv.org/abs/2203.10716
 Januschowski2019 Januschowski T, Gasthaus J, Wang Y, Salinas D, Flunkert V, BohlkeSchneider M, et al. Criteria for classifying forecasting methods. Int J Forecast. 2020;36: 167–177. doi:10.1016/j.ijforecast.2019.05.008
L Ma (2022). 'The Time Series Forecasting Problem', Datumorphism, 04 April. Available at: https://datumorphism.leima.is/wiki/forecasting/forecastingproblem/.