Highlights •After collecting historical electricity consumption data, we designed calculation indicators and created an indicator system.•A novel convolutional neuron network architecture called FLB-DCNNs, which is a focal-loss based 1d densely connected convolutional networks, is proposed to detect electricity theft.•The statistic test methods are employed to verify the performance of different algorithms, and the empirical research results prove the validity of the proposed method.•Compare this review with existing state-of-the-art methods.