Smart meters play important and basic role in advanced metering infrastructure (AMI), which efficiently improve automation and intelligence of power system. However, this modernization also introduced a lot of scope for the different anomalies and attacks on smart meters. To solve the anomaly detection problem for smart meters, three swarm intelligence based anomaly detection methods are proposed, which are based on vector distance, honesty coefficient and Kullback-Leibler divergence. The proposed algorithms mark suspicious smart meters in randomly formed swarms, and make decisions after iterations. Experimental results on real-world dataset demonstrate that the proposed algorithms are of high detection rate and low false alarm rate, which are highly practicable.