短期负荷预测的解耦决策树新算法

New Algorithm of Short-term Load Forecasting According to Decision Tree and Decoupling

短期负荷预测是电力系统调度的重要工作之一,但影响因素众多,完全由算法形成决策树易误判。为提高精度,结合决策树和解耦法将负荷预测分解为标幺曲线和平均负荷预测,根据不同预测条件分别对预测日的标幺曲线和平均负荷形成决策树,决策树前两层由实际经验指定,其余节点自动形成。充分考虑影响负荷的不同因素,越重要的影响因素越靠近决策树上层,能适应各种情况下的负荷预测。在我国北方某市的实际应用表明,该法预测精度较高。

Short-term load forecasting is an important foundational work for the power system dispatching. Because there were many influence factors, it was easy to produce misjudgment if the form of decision tree depended entirely on a algorithm. A new algorithm for load forecasting according to decision tree and deeoupling was proposed. Short-term load forecasting was divided into the per unit curve forecasting and average load forecasting to improve precision. A new method of decision tree was raised. The layers before of decision tree were designated according to experience, and the rest was formed automatically. According to different forecasting conditions, it formed the decision trees of per unit curve forecasting and average load forecasting respectively with different forecasting formulas. The new algorithm took fully into account the discipline of different factors that influenced power load and the more important factors were near the top of tree, so it was suitable for short term load forecasting in all circumstances. The practical application in one northern city shows that this method has a higher forecast precision.