关键词: 科研绩效, 核偏最小二乘法, 均方根误差, 粒子群算法

Abstract: To resolve the problem that university teachers’ performance of science research is difficult to predict, firstly, the optimization target of model is built and PSO-KPLS algorithm is addressed, which base idea is that particle swarm optimization (PSO) algorithm is used as the optimization algorithm and root mean square error (RMSE) is used as the convergence criterion, which are utilized to optimize kernel based partial least squares（KPLS） . It is used to instead of finding suitable parameter by manual operation. Then, the multi-dimensional feature vector is used to express teachers’ research performance, and the comprehensive method is used to calculate their research score. Finally, taking the data of 6 years of 60 teachers as samples, the model is trained by PSO-KPLS and fitted, and the influence of accuracy requirements on the efficiency of PSO-KPLS is mainly explored. Through the comparative experiment with other optimized KPLS algorithms, the results show that PSO-KPLS can accurately predict teachers’ research performance in the next two or three years.

Key words: research performance, kernel-based partial least-squares, RMSE, particle swarm optimization