基于改进型降噪自动编码器的家用负荷辨识方法

Household load identification method based on improvedde-noising automatic encoder

家用负荷辨识准确性受数据采样速率制约显著,过高的采样速率能够解决数据问题,但也带来成本提高、系统设计复杂等问题。基于此,提出了一种仅依赖常规采样速率有功功率量测的非侵入式负荷辨识方法,所提方法对传统的降噪自动编码器算法滑动窗的重叠部分计算进行了改进,使用中值滤波器对重叠窗的数据结果进行处理,能够较好地克服辨识结果偏高的问题。通过在REDD(reference energy disaggregation dataset)和TraceBase两个家庭用电数据集开展测试,证明了所提方法在辨识设备功率和判断设备所处状态两个方面都具有较好的效果,且各项指标均好于经典的基于因子隐马尔可夫模型(factorial hidden Markov model,FHMM)算法。另外所提算法的通用性较好,能够对不同型号、品牌的同种设备进行有效辨识,具有较好的实用价值。

The accuracy of household load identification is significantly restricted by the data sampling rate,and the high sampling rate can solve the data problem,but also bring about the cost increase,complex system design and other problems.On this basis,this paper proposes a non-invasive load identification method which only relies on the conventional sampling rate active power measurement.This method improves the calculation of the overlap part of the sliding window in the conventional de-noising automatic encoder algorithm,and adopts the median filter to process the data results of overlap window,which can better overcome the problem of high identification results.Through the testing on two household electricity consumption datasets of reference energy disaggregation dataset(REDD)and TraceBase,it is shown that the proposed method has good effect in identifying the equipment power and judging the equipment status,and its indices are better than the classical factor Hidden Markov model(FHMM)algorithm.In addition,the proposed algorithm has good versatility and can effectively identify the same equipment of different models and brands,which has good practical value.