Abstract: The primary frequency modulation performance of thermal power generating units is critical to the power quality and safe and stable operation of the power grid. Based on the primary frequency modulation assessment indicators, it is urgent to improve the primary frequency modulation performance of thermal power generating units in the west Inner Mongolia power grid. Through the analysis of the definitions, formulas and actual assessment results of a frequency modulation assessment index, the assessment calculation formulas for 15 s output response index and 30 s output response index are proposed to provide more reasonable modification suggestions under specific scenarios. By using extreme learning algorithm, this paper builds the neural network model, establishes a satisfactory non-linear mapping between the primary frequency modulation assessment index and the influencing factors affecting the primary frequency modulation performance. On this basis, according to the variation direction and degree of the characteristic variable, this paper carries out reverse analysis to analyze the reason for the failure of the primary frequency modulation performance assessment index, and then propose targeted improvement and promotion strategies, which can directly guide the optimizing work.