基于随机森林的变压器油温预测方法

Prediction Method of Transformer Oil Temperature Based on Random Forest Method

摘要: 变压器的油温可以有效反映电力变压器的工况，但要预测特定用户区域的未来需求是困难的，因为它随工作日、假日、季节、天气、温度等的不同因素变化而变化。现有预测方法不能适用于长期真实世界数据的高精度长期预测，管理人员不得不根据经验值做出决策，而经验值的阈值通常远高于实际需求而导致浪费，且任何错误的预测都可能产生严重的后果，因此需要一种有效的方法来预测未来的用电量。随机森林(Random Forest，简称RF)是Bagging的一个扩展变体，其原理简单、容易实现、计算开销小，但又具有强大的性能，代表目前最先进的集成学习技术水平的方法。本文通过收集到的变压器数据集，利用随机森林回归的预测方法，对变压器的油温变化进行预测。

Abstract: The oil temperature of a transformer can effectively reflect the working condition of a power transformer, but it is difficult to predict the future demand of a specific user’s area because it varies with different factors such as weekdays, holidays, seasons, weather, and temperature. The existing forecasting methods cannot be applied to high-precision and long-term forecasting of long-term real-world data, and managers have to make decisions based on empirical values, which are usually much higher than the actual demand and lead to waste, and any wrong prediction may result in serious consequences, so an effective method is needed to forecast future electricity consumption. Random Forest (RF) is an extended variant of Bagging with simple principles, easy implementation, low computational overhead, yet powerful performance, and is a method representing the state of the art in integrated learning. In this paper, we use the prediction method of Random Forest regression to predict the oil temperature variation of transformers from the collected transformer data set.