The Application of Energy Consumption Prediction Model for Fresh Air System based on PSO-BP and BO-BP

With the fast progress of deep learning and neural network technology, neural networks are now being used in lots of different subjects. They have been especially popular in predicting energy use in fresh air systems, because of their ability to fit nonlinear data. However, the performance of neural networks largely depends on the selection of their hyperparameters, and how to effectively optimize these hyperparameters has become a key issue to improve the prediction accuracy of the model. This paper discusses the methods of optimizing the hyperparameters of neural networks and compares several optimization methods, and finally finds that the Bayesian optimization algorithm performs optimally in neural network optimization. In the experiment, firstly, the relevant data of the new wind system were collected and data preprocessing was carried out, including missing value interpolation and outlier detection. The key factors affecting the energy consumption of the fresh air system were then used to train a prediction model. Then, PSO and BO were used to improve the neural network's hyperparameters. A Monte Carlo simulation showed the Bayesian algorithm identifies the global optimum more quickly, improving prediction accuracy. Experimental findings show that factors like indoor-outdoor temperature, humidity and air quality influence the energy consumption of fresh air systems.The BP neural network model, optimised by the Bayesian optimisation algorithm, performs better when predicting energy consumption. Research results offer an effective energy consumption prediction method for fresh air systems and important experimental support for related energy prediction tasks.