Electric utility companies around the world have suffered significant financial losses as a result of the urgent problem of electricity theft. Every year, electricity worth $6 billion is stolen in the United States alone. Conventionally, physical attacks such as line tapping or meter manipulation are used to steal electricity in the consumption domain. Attacks involving electricity theft can take on new shapes thanks to the smart grid paradigm. In order to detect electricity cyber attacks, The performance of a number of deep learning algorithms is evaluated in this paper, which include convolutional neural networks (CNN), deep feedforward neural networks (DNN), and recurrent neural networks with gated recurrent units (RNN-GRU). Electricity is produced by solar panels in modern, developed countries. These customers will receive two different meters that measure production and consumption data, and they can sell their extra energy to those in need. When producing, some unscrupulous users could tamper with smart meters to increase the amount of money that can be obtained from distributed renewable energy sources and power. This attack can cost agencies a great deal of money. This initiative uses deep learning algorithms, which can identify any change and predict theft, to detect such attacks.