To improve the evaluation capability of the flood risk assessment model in Xinxiang City, six methods including analytic hierarchy process(AHP), logistic regression (LR) model, BP neural network, random forest (RF) model, and PSOBP model and PSORF model combined with metaheuristic algorithm particle swarm optimization (PSO) were used to conduct flood risk assessment in Xinxiang City, generating a flood inventory map containing 200 flood locations. Nine flood impact factors were selected, and the correlation between flood impact factors and flood occurrence was analyzed using variance inflation factor. The evaluation capabilities of six flood risk assessment methods were compared using confusion matrix and subject working characteristic curve, and finally obtaining the flood risk distribution maps of the six methods. The results show that the evaluation performance of PSORF and PSOBP models is better than that of single algorithms, and the area under curve of the receiver operating characteristic curve is 0.953 and 0.947, respectively. According to the obtained flood risk distribution map, at least 36.5% of the areas in Xinxiang City are classified as highly susceptible to flood impacts, and the flood risk assessment model coupled with metaheuristic algorithm has higher accuracy.