To prevent accidents caused by the hidden danger of hydraulic steel gates and to avoid the waste of manpower and material resources in overhaul and maintenance process,it is necessary to evaluate the safety grade of the hydraulic steel gate.At present,there are many models for the safety grade evaluation of hydraulic steel gates at home and abroad,but most of them are only applicable to the traditional periodic inspection,while few models for online safety grade evaluation system .The hydraulic steel gate safety ratings can be abstracted as a pattern recognition problem,and the BP neural network is widely used in various fields of pattern recognition because of the strong learning ability,ability to adapt,and fault tolerance.Thus BP neural network can be constructed for the security level of hydraulic steel gate recognition model.Therefore,exploring the neural network for the safety of hydraulic steel gate online evaluation system is feasible.There are too many evaluation indexes of the traditional hydraulic steel gate safety grade,which leads to the redundancy of the input feature vector dimension.Moreover,if the initial value of the BP neural network is not good,it is easy to cause the training result to fall into the local optimal.So,the accuracy of BP neural network for the identification of the safety grade of hydraulic steel gates is not ideal. Since the information gain (IG) can achieve the quantification of the safety level of hydraulic steel gate and the correlation between each feature.The self-adaptive flower pollination algorithm (SFPA) has a strong global search ability and a local search ability and can realize the optimization of the initial value of BP neural network.Given this,an IG-SPAF-BP neural network model is proposed.According to the change of entropy,the information gain of each hydraulic steel gate safety grade feature is calculated,and the features carrying more information are selected as the input features of the neural network to reduce the training time of the neural network and improve the generalization ability of the neural network.By using the global and local pollination of the SFPA algorithm,the initial weight and initial threshold of the BP neural network are optimized to further improve the convergence speed of the BP neural network and the identification accuracy of the safety grade of the hydraulic steel gate. The repeated operation results of the IG-SFPA-BP network model,the standard BP network model,the IGBP network model,the IG-FPA-BP network model,the IG-GA-BP network model,and the IG-PSO-BP network model on the hydraulic steel gate sample set were compared.The experimental results show that in the above five kinds of models,the average recognition accuracy of the IG-SFPA-BP network model is the highest.However,the average running time of the IG-SFPA-BP network model is longer than that of the standard BP network model,IGBP network model,and IG-GA-BP network model,and compared with IG-GA-BP network model and IG-PSO-BP network running time is shorter.Besides,the mean square error of the IG-SFPA-BP network model is the smallest and the error curve is the most stable. The information gain theory reduces the dimension of the feature vector for the identification of the safety grade of hydraulic steel gate,shortening the running time of BP network and improving the classification ability of the network to a certain extent.SPFA algorithm can significantly improve the accuracy of the BP neural network in identifying the safety level of the hydraulic steel gate.IG-SFPA-BP model has good applicability for the identification of the safety grade of the hydraulic steel gate.According to the comparative test,its learning ability,generalization ability,and stability are better than those of standard BP,IG-BP,IG-FPA-BP,IG-GA-BP,and IG-PSO-BP network models.It can provide algorithm support for an online evaluation system of safety grade of hydraulic steel gate,and provide a reference for the followup maintenance of gate.Although the IG-SFPA-BP model has a good performance for the identification of the safety level of hydraulic steel gates,the number of samples collected is small,and some samples have vacancy characteristics.Therefore,the identification accuracy of the model has a large space for improvement,and the sample data set needs to be further improved. Further research is needed to determine the number of hidden layer neurons in the BP neural network.