基于卷积神经网络的特高压直流线路保护动作行为评价

Evaluation of UHV DC line protection action behavior based on convolution neural network

特高压直流(ultra hight voltage direct current,UHV DC)系统控保装置的可靠性对电力系统的安全稳定运行至关重要。目前特高压直流线路保护存在可靠性不高的问题,准确迅速地评价特高压直流线路保护的动作行为,对于提高电力系统的安全稳定运行具有重要意义。提出了一种利用卷积神经网络对特高压直流线路保护动作行为进行评价的方法,首先分析了影响线路保护动作行为的因素;然后构建了评价线路保护动作行为的卷积神经网络结构,开展了基于PSCAD(power systems computer aided design)的直流线路保护动作行为仿真,生成了卷积神经网络的训练和测试样本集,再利用训练样本集对评价保护动作的卷积神经网络进行训练;最后对该网络进行了测试,测试结果表明,该方法可以正确评价特高压直流线路保护的动作行为,通过与传统人工网络的评价结果的对比可知,该方法具有速度快、错误率低、学习能力强等优点。

The reliability of the control and protection device of the ultra hight voltage direct current(UHV DC)system is crucial to the safe and stable operation of the power system.At present,the UHV DC line protection has the problem of low reliability. Evaluating the action behavior of UHV DC line protection accurately and quickly plays an important role in improving the safe and stable operation of power systems.This paper presents a method to evaluate the protection action behavior of UHV DC lines using convolution neural networks.First of all,the factors affecting the action behavior of line protection are analyzed.Then, a convolution neural network structure is constructed to evaluate the action behavior of line protection.The simulation of action behavior of DC line protection based on power systems computer aided design(PSCAD) is carried out to generate training and test sample sets of convolution neural networks. After that the training sample sets are used to train the convolution neural network to evaluate the protection action.Finally, the network is tested.The test results show that this method can correctly evaluate the protection action of UHV DC lines.Compared with the evaluation results of the traditional artificial network,this method has the advantages of fast speed,low error rate and strong learning ability.