The Research And Application Of Improved BP Algorithm In Monitoring And Prediction Of Seawall Multipoint Osmotic Pressure

The economy of coastal areas developing rapidly, in order to ensure the security ofpeople’s life and property behind the seawall, its importance is taken seriously more andmore, and that related departments have strengthened the protection of seawall. Seawallis built along the coast, long and wide-ranging, its work environment is very complex,prone to security risks, therefore, strengthen the seawall safety monitoring is essentialand significant. Along with the advancement of seawall construction management work,seawall levee’s safety monitoring has gradually become an important means to ensurethe safety and operability of the seawall, gaining more and more attention. Osmoticpressure is a major indicators affecting the safety of seawall, so osmotic pressuremonitoring plays an important role in the analysis of seawall safety monitoring.BP neural network is a relatively mature network, which has strong nonlinearmapping ability. Considering the complexity of the seawall work environment, as wellas the ambiguity between the osmotic pressure and its influence factors, this paper triesto apply the BP neural network to seawall’s osmotic pressure monitoring and prediction.The standard BP algorithm based on gradient descent method exists deficiencies inapplication, by analysis of the shortcomings of the algorithm, we selects the optimalactivation function method and the additional momentum algorithm to improve thestandard BP neural network respectively. Activation function optimization method is toadd adjustable parameters in the activation function in the formula, adjusting functiongradient and the range of mapping to achieve the purpose of improving the performanceof the network; additional momentum algorithm is based on the value of each weightplus a variation that proportional to the previous weights, thereby accelerating networkweights updated to improve standard BP algorithm. The previous two methods are fromtwo different angles to improve standard BP neural network, on this basis, we propose acombination improved algorithm, which combined additional momentum algorithmwith activation function optimization.Based on the measured data of Pudong seawall, considering that single pointmodeling not only big workload, but also that the information correlation betweenseepage pressure measuring points is not high, this article revise from the perspective ofmultiple measuring points for modeling, analyzing the overall impact of the tide,rainfall, aging and other factors on the osmotic pressure of the seawall. Calculate the simplified factors and combination of factors by means of BP neural network, to select agroup which has better prediction results as the input.After the network structure is determined, make the three improved BP algorithmprogramming models separately, than analyze their osmotic pressure monitoringmultiple measuring points in the application of the seawall, and compare the improvedmodel for predicting the effect of osmotic pressure seawall according to the training andprediction results. The results showed that the three improved BP neural network haveimproved in terms of speed and accuracy, where a combination of improved model hasbetter prediction accuracy than a single improved model, and achieved good results inthe analysis of the seawall seepage pressure monitoring forecasting model.