Combined Predication Of Relic Disease

Open-air cultural relics have important historical values,and have suffered varying degrees of damages due to the atmospheric environment and the deterioration of the natural environment,And with the development of the IoT and sensors,some open-air relics diseases and the collection of environment information have realized real-time detection and network sharing,Because of the lack of quantitative analysis in traditional artificial forecasting methods,it is more important to predict the risk of cultural relics diseases quantitatively.A Corr-BP neural network based on correlation is designed in this paper,and dragonfly algorithm is used to improve its origin weights and cost function,which solves the problems of low predication accuracy and overfitting in traditional BP neural network prediction method,then the relic disease combined predication model is built.The main works are as follows:Firstly,for the overfitting problem exists in traditional BP neural network,the cost function of which is improved,and the Pearson correlation was introduced as the cost function to establish a combined Corr-BP neural network,whose structural parameters,training methods and model optimization are analyzed.The experiments on standard datasets show the Corr-BP has the capability to solve the overfitting problem and can improve the predication accuracy.Then,the MODA-Corr-BP neural network model combining with the Dragonfly algorithm is proposed.By modifying the cost function of Corr-BP neural network to multi-objective function the initial distribution of weights is improved.The experiments prove that the proposed model improves the Corr-BP's stability and has higher forecasting accuracy and training efficiency than BP and RBF neural network.Finally,the ARIMA and grey forecasting model are combined with MODA-Corr-BP neural network model respectively in the short and medium predication of the cracks and displacements in the stone carvings of Tangshunling,and the results show that the proposed model for relic disease prediction can effectively decrease the predication error and can satisfy the demand for relic disease predication.