A hybrid neural network model based on Wavelet and LSTM is proposed to restrict the anomaly power consumption behavior in transmission and distribution system. Firstly, an abnormal power consumption simulation algorithm is proposed to generate abnormal power consumption data sequence. Next, the feature extracting network which is constructed by using LSTM network extracts different sequence features from power consumption data. Lastly, the mode mapping network with wavelet neural network as the core, uses the extracted different sequence features to detect the anomalous electrical power consumptions. Case studies on CER Smart Metering Project datasets have demonstrated that the proposed model has higher detection rate, lower false positive rate and higher Bayesian detection rate, compared with conventional networks.