Energy Consumption Prediction of Chillers Based on Improved Gradient Descent Algorithms

Low-carbon energy transition is currently a key development strategy for China. In the total energy consumption of society, the energy consumption of cooling systems accounts for a significant proportion. Therefore, precise energy consumption prediction, as the first step of optimization, is crucial. This paper proposes a Back Propagation Neural Network model trained with an improved gradient descent algorithm, which combines the Horizontal Cross Particle Swarm Optimization (HCPSO) algorithm with the gradient descent algorithm. The HCPSO introduces a crossover operation based on the traditional Particle Swarm Optimization, further enhancing the global search capability. The improved algorithm is applied to train the energy consumption of chiller systems within the cooling source system, demonstrates faster convergence and higher prediction accuracy compared to the traditional gradient descent training method, which is of practical significance for the modeling and optimization of energy consumption in cooling systems.