Deep learning-based data centre energy efficiency analysis and optimisation research

With the rapid development of the digital economy, data centres, as the core carrier of the arithmetic infrastructure, are facing increasingly severe energy consumption problems. The traditional data centre operation is based on resource redundancy and fixed strategies, which is difficult to adapt to the demand of dynamic load and energy efficient synergy, and there is an urgent need to build an intelligent and green computing support system to achieve the optimal control of energy efficiency under the goal of ‘carbon peak and carbon neutrality’. This study focuses on the theme of ‘green computing in data centres’, and adopts data-driven modelling, deep learning prediction and other methods to systematically construct a set of energy-efficiency modelling and scheduling optimization frameworks oriented to the synergy of arithmetic power, cooling and energy consumption. The study firstly analyses the temporal characteristics of heterogeneous data from multiple sources in the data centre, and constructs a multi-model fusion energy consumption prediction system based on neural networks such as deep neural networks (DNN), Long Short-Term Memory Network (LSTM) and Transformer. On this basis, a green energy-efficiency oriented decision-making mechanism is established by introducing a multi-objective optimisation method, taking into account the indicators of energy consumption, PUE, carbon emission and response delay. The empirical results show that the proposed green computing technology scheme can reduce Power Usage Effectiveness (PUE) and overall energy consumption in a typical data centre, significantly improve resource utilization efficiency and green operation level, and provide an effective technical path and demonstration support for the intelligent and low-carbon transformation of China's data centres.