The big data sources in smart grid (SG) enable utilities to monitor, control, and manage the energy system effectively, which is also promising to advance the efficiency, reliability, and sustainability of energy usage. However, false data attacks, as a major threat with wide targets and severe impacts, have exposed the SG systems to a large variety of security issues. To detect this threat effectively, several machine learning (ML)-based methods have been developed in the past few years. In this paper, we provide a comprehensive survey of these advances. The paper starts by providing a brief overview of SG architecture and its data sources. Moreover, the categories of false data attacks followed by data security requirements are introduced. Then, the recent ML-based detection techniques are summarized by grouping them into three major detection scenarios: non-technical losses, state estimation, and load forecasting. At last, we further investigate the potential research directions at the end of the paper, considering the deficiencies of current ML-based mechanisms. Specifically, we discuss intrusion detection against adversarial attacks, collaborative and decentralized detection framework, detection with privacy preservation, and some potential advanced ML techniques.

智能电网（SG）中的大数据源使公用事业可以有效地监视，控制和管理能源系统，这也有望提高能源使用的效率，可靠性和可持续性。然而，错误数据攻击作为具有广泛目标和严重影响的主要威胁，已使SG系统面临各种安全问题。为了有效地检测到这种威胁，在过去几年中已经开发了几种基于机器学习（ML）的方法。在本文中，我们对这些进展进行了全面的调查。本文首先简要介绍了SG体系结构及其数据源。此外，还介绍了错误数据攻击的类别以及随后的数据安全性要求。然后，最近的基于ML的检测技术归纳为以下三种主要检测方案：非技术损失，状态估计和负荷预测。最后，考虑到当前基于机器学习的机制的不足，我们在本文的结尾进一步研究了潜在的研究方向。具体来说，我们讨论了针对对抗性攻击的入侵检测，协作式和分散式检测框架，具有隐私保护的检测以及一些潜在的高级ML技术。