Prediction of soil thermal conductivity using Zebra algorithm-optimized machine learning model

Purpose This study constructed a training set with 257 sets of STC measurement data and considered six influencing factors as input. The research uses ZOA to optimize the prediction model of the BP neural network. Sensitivity analysis on the input amount is conducted using the cosine amplitude method. Then, the ZOA–BP model was used to predict STC. In addition, in order to assess the predictive accuracy of the ZOA–BP model, the prediction results of five classic models, SPE, DELM, RF, BP and SVR, were used as the control group. At the same time, RMSE, MAPE, R2 and a10-index are used as evaluation indicators to evaluate the prediction performance of the model comprehensively. Design/methodology/approach This study uses a machine learning hybrid prediction model to predict STC under different influencing conditions. A data set of 257 sets of data is selected for prediction. First, the BP neural network was optimized based on the ZOA, the STC prediction model of ZOA-BP was developed and CAM was used to perform sensitivity analysis on the input parameters. In addition, single models were used as the control group and the prediction effect was scored through RMSE, R2, MAPE and a10-index. Findings (1) This hybrid drive model can meet the actual needs of STC prediction. (2) The RMSE, MAPE, R2 and a10-index of the BP neural network model optimized by ZOA are all better than the single models in the control group. The prediction performance of the six models is as follows: ZOA–BP > RF > SVR > DELM > BP neural network > SPE. (3) Machine learning models have more significant advantages in prediction accuracy than empirical formulas. (4) The importance of the six input parameters to STC prediction is clay soil content > saturation degree > sand content > porosity > quartz content > soil dry density. Originality/value We hereby confirm that this manuscript is our original work and has not been published nor has it been submitted simultaneously elsewhere. We further confirm that all authors have checked the manuscript and have agreed to the submission. All the figures and tables are authors’ own work.