Multi-dimensional prediction and factor analysis of thermal performance for energy piles

Energy piles operate in a complex and various environment, and its thermal performance is affected by many factors. To accurately predict the thermal performance and explore the sensitivity of factors, we introduce a novel hybrid model, and directly and quantitatively evaluate the primary and secondary influencing factors of thermal performance from the perspective of causal relationships. First, the principal component analysis (PCA) reduces eight influencing factor data of energy piles into four principal component data, obtaining data with different dimensions. Secondly, through the particle swarm optimization (PSO) iteration, the optimal Gaussian kernel function is determined among four kernel functions. Then, the relevance vector machine (RVM) is used to predict the thermal performance by six-fold cross validation method. Compared with the predicted results of eight-factor data, the prediction results of four principal component data have a higher accuracy, and the mean relative error (MRE), root mean square error (RMSE), and Thiel's inequality coefficient (TIC) are 4.776 %, 0.969, and 0.494 %, respectively. Finally, through the sensitivity analysis of factors, we provide the specific impact values and sensitivity rankings of influencing factors, and use three radar charts to visually present the sensitivity distribution of factors.