多步负荷预测

Multi-step load forecasting

  在说Multi-step之前，先说下one-step。
  one-step是指预测下一个时间步的值，可表示为：$$P(t)=model(y(t-1),y(t-2),\dots ,y(t-n))\text{\hspace{1em}(1)}$$式中$P(t)$为时间步t的预测值。
  Multi-step则是指预测多个连续时间步的值，有4种常见形式[1]：- Direct multi-step forecasting- Recursive multi-step forecasting- Direct-Recursive hybrid multi-step forecasting- Multiple output forcasting

Direct Multi-step Forecasting

  一个模型预测一个时间步，可表示为：$$P(t)=mode{l}_1(y(t-1),y(t-2),\dots ,y(t-n))\text{\hspace{1em}(2.1)}$$ $$P(t+1)=mode{l}_2(y(t),y(t-1),\dots ,y(t-n))\text{\hspace{1em}(2.2)}$$  这样的缺点是，忽略了预测结果之间的相关性。

Recursive Multi-step Forecasting

  重复使用一个模型，将一个时间步的输出作为下一个时间步的输入，可表示为；$$P(t)=model(y(t-1),y(t-2),\dots ,y(t-n))\text{\hspace{1em}(3.1)}$$ $$P(t+1)=model(P(t),y(t-1),y(t-2),\dots ,y(t-n))\text{\hspace{1em}(3.2)}$$  这样的缺点是：会造成误差(error)的不断累积。

Direct-Recursive Hybrid Multi-step Forecasting

  结合上述两种方法，可表示为：$$P(t)=mode{l}_1(y(t-1),y(t-2),\dots ,y(t-n))\text{\hspace{1em}(4.1)}$$ $$P(t+1)=mode{l}_2(P(t),y(t-1),y(t-2),\dots ,y(t-n))\text{\hspace{1em}(4.2)}$$

Multiple Output Forcasting

  一个带有多个输出的模型。即$$P(t),P(t+1)=model(y(t-1),y(t-2),\dots ,y(t-n))\text{\hspace{1em}(5)}$$  Multiple output models are more complex as they can learn the dependence structure between inputs and outputs as well as between outputs[1].

[1] 4 Strategies for Multi-Step Time Series Forecasting