【摘要】 电力时序数据的分类问题是数据挖掘领域中的一个研究热点。对电力用户模式分类可以帮助电网企业分析用户用电特性,实现差异化营销。在回声状态神经网络(echo state network,ESN)基础上引入过程神经元,提出一种用于电力时序数据分类的新型神经网络——过程回声状态网络(process echo state network,PESN)。该网络通过对连接权重进行函数化,实现函数域到实数域的映射。并针对权函数难以训练学习的问题,提出利用基函数的展开形式来逼近权函数的学习算法,从而将回声状态神经网络从时序预测领域进一步扩展到时序分类领域中。仿真实验结果证明,与传统时间序列分类算法相比,过程回声状态网络在用电模式分类实验中表现出了更好的效果。更多还原

【Abstract】 Time series classification is a hotspot of data mining research. Classifying electricity customers can help power grid enterprises analyze customers’ patterns of electricity consumption and achieve differentiated marketing. By introducing process neurons, a process echo state network(PESN) is proposed based on echo state network(ESN) for time series classification. Because connection weights of process neurons can be time-varying functions, PESN realizes mapping from function field to real number field. In order to solve problem of learning weight functions, an algorithm approximating the weight functions with basis functions expansion is proposed. Consequently, PESN expands application domain of ESN from time series prediction to time series classification. Compared with other time series classification algorithms, PESN shows better results in electricity customers’ classification.更多还原