Prediction and Optimization on Energy Consumption of Data Center Based on Multi-layer Feedforward Neural Network

摘要

There will be the characteristics of large amounts of equipments, high parameter coupling, non-linear calculation and complex modeling for the conventional energy consumption calculation. Multi-layer feedforward neural network model is used to establish relations between the parameters of air conditioning system, computer equipments, power supply system and the energy consumption values. The error back-propagation algorithm based on gradient descent strategy is used to adjust connection weight and threshold of the neurons in hidden layers. The genetic algorithm is used on the initial weights and thresholds optimization and the search for minimum energy consumption. Through the prediction of energy consumption with the variation of uncontrollable parameters, the adjustment of controllable parameters such as the temperature target value of air conditioner, the control mode of fresh air exchangers and humidifiers can obtain the target of energy consumption minimization.