基于多尺度卷积与Informer混合模型的非侵入式负荷监测方法

Non-intrusive load monitoring method based on multi-scale convolution and Informer hybrid model

针对现有非侵入式负荷监测方法存在的负荷分解准确率低、模型泛化性能差的问题,提出一种多尺度卷积与Informer网络相结合的非侵入式负荷监测方法。采用数据分段优化方法对功率信号进行分段,利用多尺度卷积核获取不同时间尺度的特征序列以及自适应提取多维度功率特征,从而形成特征矩阵;基于Informer网络中的概率稀疏自注意力机制在高维空间中充分捕获特性序列的长期依赖关系,从而提高预测准确率;利用分解值修正方法消除功率分解值中的“虚假”激活状态,以进一步提高分解精度。算例结果验证了所提方法的可行性。

Aiming at the problems of low load decomposition accuracy and poor model generalization perfor-mance existing in the current non-intrusive load monitoring methods,a non-intrusive load monitoring method combining multi-scale convolution and Informer network is proposed.The data segmentation optimization method is adopted to segment the power signal,a multi-scale convolution kernel is used to obtain the feature sequences of different time scales and adaptively extract multi-dimensional power features,thus a feature matrix is formed.The long-term dependence relation of feature sequences in high-dimensional space is cap-tured based on the probability sparse self-attention mechanism in Informer network,thus the prediction accuracy is improved.The decomposition value correction method is used to eliminate the“spurious”activa-tion states in the power decomposition values for further improving the decomposition accuracy.The feasibility of the proposed method is verified by the example results.