基于双向GRNN与时间序列翻译模型的非侵入式负荷分解算法

Non-invasive Load Decomposition Algorithm Based on Bidirectional GRNN and Time Series Translation Model

非侵入式负荷分解是用户侧精细化能量管理的关键技术,为了提高算法分解准确率与模型训练速度,提出了一种基于双向GRNN与时间序列翻译模型的非侵入式负荷分解算法。使用局部注意力机制对中间向量的传递过程进行了优化,在增加模型注意力的同时降低了算法的运算量。使用集束搜索算法使解码环节得到了更多的功率概率曲线,实现了分解准确率的提高。使用人工合成训练数据方法克服了数据集不平衡问题,提高了算法的稳定性。最后在REDD数据集上对文章所提算法进行了验证,与其他先进算法相比,所提出算法的准确率具有较大幅度提高,并且与基于LSTM的算法相比,本算法的模型训练速度提高了40%以上。

Non-invasive load decomposition is the key technology of refined energy management on the user side.In order to improve the algorithm decomposition accuracy and model training speed,a non-invasive load decomposition algorithm based on bidirectional GRNN and time series translation model is proposed.The transfer process of intermediate vector is optimized by using the local attention mechanism,which increases the attention of the model and reduces the computational burden of the algorithm. Using the cluster search algorithm,more power probability curves are obtained in the decoding segment,and the decomposition accuracy is improved.The method of artificial synthesis training data overcomes the problem of unbalanced data sets and improves the stability of the algorithm.Finally,the algorithm proposed in this paper is verified on the REDD data set.Compared with other advanced algorithms,the accuracy of the algorithm proposed in this paper is greatly improved,and the model training speed of this algorithm is increased by more than 40 percent compared with the algorithm based on LSTM.