An energy consumption monitoring method based on neural network transparency

The invention provides an energy consumption monitoring method based on neural network transparency, which comprises the following steps of: generating a neural network establishment model according to production data; Generating a neural network paraphrasing graph; Obtaining connection weight index data; And carrying out significance test on the connection weight according to the weight receivingindex data and deleting non-significant connection weight in the neural network paraphrasing graph. According to the method, a neural network paraphrasing graph, a connection weight method and an improved randomization test method are combined, a very good way is provided for transparency of a neural network model in a complex industrial process, and an effective guidance basis is provided for energy consumption analysis of enterprise production data. 本发明提供种基于神经网络透明化的能耗监测方法，包括以下步骤，根据生产数据生成神经网络建立模型；生成神经网络释义图；获取连接权指标数据；依据接权指标数据对连接权进行显著性检验并删除在神经网络释义图中不显著的连接权。本发明结合神经网络释义图、连接权法和改进的随机化测验三种方法，为复杂工业过程神经网络模型的透明化提供了种很好的途径，为企业生