Analytic Hierarchy Process (AHP) is an effective method to evaluate complex decision making processes, but the weights are always manually assigned. Therefore, the neural network method is usually used to distribute the weights. Thus, the subjective errors of manual allocation can be avoided. However, traditional neural networks tend to fall into local extremes and cannot find optimal solutions. At the same time, irrelevant variables and too long historical data will also affect the results of the assessment. Therefore, this paper proposes a new deep residual network analytic method combined with multiway principal component analysis (MPIDRN-AHP). First, all the historical data are divided into several groups. Each group performs principal component analysis, extracts the operation variables related to the target variable, and then all the selected operation variables are sorted by frequency. The highest frequency operation variables are selected to put into the constructed network. Using deep network, the proposed MPIDRN-AHP can obtain the optimal solution of AHP global weight, which can not only reduce the subjective error but also avoid the gradient disappearance problem caused by the increase of the network layer. The proposed method is used in the evaluation of TEP. Compared with the traditional AHP, BP-AHP and DRN-AHP, the proposed MPIDRN-AHP method can achieve the most satisfactory performance.