基于RF-BiLSTM的柔直阀冷入阀水温预测及冷却能力评估

为实现柔性直流(voltage sourced converter-high voltage direct current,VSC-HVDC)换流阀冷却系统入阀水温的智能预测，文中提出一种基于随机森林(random forest,RF)和双向长短时记忆(bi-directional long short-term memory,BiLSTM)网络混合的柔直换流阀冷却系统入阀水温的预测模型，并以此为基础对柔直换流站阀冷系统的冷却能力进行评估。首先，采用RF算法对由阀冷系统监测变量组成的高维特征集进行重要性分析，筛选出影响入阀水温的重要特征，与历史入阀水温构成输入特征向量。然后，将特征向量输入到BiLSTM预测模型，对模型进行训练并实现对入阀水温的准确预测和冷却能力定量评估。最后，以广东电网某柔直换流站为实例对所提方法进行分析，验证了所提出的基于RF-BiLSTM的混合模型预测精度优于BiLSTM模型、RF模型、支持向量机(support vector machine,SVM)模型和自回归滑动平均模型(auto-regressive and moving average,ARMA)模型，并且实现了冷却能力的定量评估。结果表明该换流站冷却裕量达98%，存在过度冷却、能源浪费的问题，与换流站现场运行情况相符，验证了文中所提方法的有效性和准确性。

Prediction of valve inlet water temperature and cooling evaluation of VSC-HVDC convertvalve cooling system based on random forest and bi-directional long short-term memory

In order to realize the intelligent prediction of valve inlet water temperature of a voltage sourced converter-high voltage direct current (VSC-HVDC) valve cooling system,a prediction model of inlet water temperature of VSC-HVDC based on a hybrid model of the random forest (RF) and bi-directional long short-term memory (BiLSTM) is proposed,and the cooling capacity of the cooling system is evaluated on the basis of the proposed prediction model. Firstly,a RF algorithm is used to analyze the importance of high-dimensional feature sets,which consist of all the monitoring variables of the valve cooling system. Then the important characteristic parameters affecting the inlet water temperature are filtered out to form an input feature vector with the historical inlet water temperature. Secondly,the feature vector is input to the developed BiLSTM prediction model to train the model for accurately predicting inlet valve water temperature and quantitatively evaluating the cooling capacity. Finally,a VSC-HVDC converter station in Guangdong power grid is taken as an example to verify the effectiveness and superiority of the proposed method. Simulation results indicate that the accuracy of the proposed hybrid model based on RF-BiLSTM is higher than that based on BiLSTM model,RF model,support vector machine (SVM) model and auto-regressive and moving average (ARMA) model. Moreover,the cooling capacity is evaluated quantitatively and accurately. Analysis results show that the cooling margin of this converter station is up to 98%,which indicates that there is a problem of overcooling and energy waste. The evaluation result of the cooling capacity is well consistent with the field operation result of the converter station,which confirms the effectiveness and the accuracy of the proposed method.