In order to comprehensively measure the randomness and ambiguity in the most stringent water resources management evaluation process, the normal cloud model was introduced into the most stringent water resources management evaluation, and the optimal foraging algorithm-projection pursuit-normal cloud evaluation model was established. Examples of the most stringent water resources management evaluations in 16 cities in Yunnan province are taken for case studies. Six indicators to build evaluation index system, such as the most stringent water resources management assessment of water consumption of 10,000 yuan GDP, and grading standards were selected, and the cloud model forward generator was adopted to calculate the membership degree of the most stringent water resources management grading evaluation index. The optimal foraging algorithm- projection pursuit method was used to work out the weight of each index, and compared with the traditional particle swarm optimization, artificial bee colony algorithm and differential evolution algorithm optimization results. The degree of certainty of the most stringent evaluation of water resources management is given and evaluated according to membership matrix and weight matrix. The results show that the precision of the optimal foraging algorithm is higher than that of the traditional particle swarm optimization algorithm, the most stringent water resources management evaluation of Kunming and Qujing City was excellent, Baoshan City, Honghe Prefecture, Dehong Prefecture were evaluated as qualified, the remaining 11 states were rated as good. The optimal foraging algorithm-projection pursuit-normal cloud evaluation model has both fuzziness and randomness, which not only reflect the qualitative concept of the most stringent evaluation and classification of water resources management, but also reflect the uncertainty of the degree of membership, which is of good promotional value.