With the widespread deployment of artificial intelligence (AI) applications and the rapid expansion of digital infrastructure, data centers have experienced continuous growth in scale and computing demand, making energy consumption management a critical issue in the energy and power engineering domain. Distinct from conventional electricity consumers, data center loads exhibit unique operational characteristics, with total energy consumption dominated by information technology (IT) equipment and cooling systems. The strong non-linearity, multi-source coupling, and dynamic evolution of these components pose significant challenges to make accurate energy consumption forecasting and power system dispatch. The paper systematically reviews the research progress in data center power supply configurations, load modelling, and energy consumption forecasting. Particular emphasis is focused on a comparative analysis of statistical models, machine learning, and deep learning, which are applied to IT load and cooling energy prediction, applicable conditions, feature construction strategies, and modeling performance, etc. Based on extensive literature analysis, key feature selection principles and model adaptation patterns under multi-source monitoring data are summarized, highlight the role of workload characteristics, thermal environmental variables, and operational control parameters to improving prediction. Furthermore, data center participation mechanisms in power system operation, dispatch based on load and energy consumption forecasting are reviewed also. From both temporal and spatial perspectives, the influence of forecasting on demand response implementation, flexible regulation capability extraction, and coordinated operation between power system and computing resource is analyzed. The results indicate that high-resolution, reliable load forecasting is a prerequisite for unlocking the flexibility potential of data center and enhancing power-computing synergy in modern power system. Finally, considering practical deployment requirements, the paper identifies key challenges faced by current research, including insufficient multi-source data fusion, limited model interpretability, and constrains in real-time forecasting and decision-making. Cutting-edge and future research directions and implementation strategies are proposed to support intelligent energy management and coordinated power-computing operation of data centers, providing systematic references and methodological insights for researchers and practitioners in this field.