With the strengthening of environmental protection awareness, constructed wetlands are an important part of urban ecosystems, and their environmental status assessment is paid more attention. This study aims to construct and validate a new model for the environmental state assessment of constructed wetlands, which integrates the analytic hierarchy process and backpropagation neural network to improve the accuracy and efficiency of assessment. This study collects a large amount of wetland environmental data, uses the analytic hierarchy process to determine the weights of key evaluation indicators, and then uses an improved backpropagation neural network to carry out in-depth learning and prediction to accurately assess the environmental state of constructed wetlands. An adaptive variation genetic algorithm was used to optimize the BP neural network model, along with L1 and L2 regularization techniques and Adam and RMSprop optimization algorithms were used to further improve the model performance. The results show that the model converges at the 20^th iteration, the convergence speed is increased by 34%, 80% of the data falls within the error range of plus or minus 0.2, the R2 value is as high as 0.99163, and the mean square error is close to zero, which shows that the model has significant advantages in improving the evaluation accuracy and efficiency. The comprehensive score shows that the overall environmental status of the constructed wetland is not ideal; there may be pollution problems, and improvement measures need to be taken. This model provides a scientific basis for wetland environmental protection and management, helps optimize protection measures, and realizes the sustainable development of wetland ecosystems. It is also of great significance for improving the efficiency and effect of wetland management.