Among the many environmental factors in the cold chain storage and transportation environment, the temperature and humidity in the cabin are the key factors, and they have characteristics such as nonlinearity and strong coupling. At the same time, the acquired data has noise interference, In order to solve the traditional single-point forecast cannot meet the needs of cold chain storage and transportation environmental performance evaluation, a multi-step prediction method for refrigerated compartment temperature and humidity based on the combination of K-medoids and long short-term memory network (LSTM) was proposed. The historical temperature and humidity data in the refrigerated compartment and the distribution characteristics of the collection nodes were taken as input according to the time series, and K-medoids were used for data fusion, and then the fused data was input into the LSTM network according to the time series for temperature and humidity prediction. The prediction method was applied to the prediction of temperature and humidity in the refrigerated vehicle of Zhoushan Xingye Group. The test results showed that the RMSE of the prediction method for the temperature in the refrigerated vehicle was 0.3438℃, the MAE was 0.2730℃, and the MAPE was 1.51%; the RMSE of the humidity in the refrigerated compartment was 2.5619%, the MAE was 1.9956%, and the MAPE was 3.53%; compared with K-medoids-BP, K-medoids-RBF, K-medoids-Elman neural network model, all showed that the proposed model had higher prediction accuracy, and can provide strategic support for the fine management and regulation of the cold chain transportation environment.