Prediction Of Ammunition Storage Reliability Based On Particle Swarm Optimization BP Neural Network

Ammunition is one important part of military equipment,and is responsible for the task of improving battle effectiveness of the army.The storage reliability of ammunition is an important indicator to evaluate the quality of ammunition,and also the core element of ammunition logistics support technology.During the storage process of ammunition,its quality will change and lead to ammunition failure due to the influence of environmental factors and human activities.The accurate evaluation and prediction of the ammunition storage reliability establish a foundation for the study of the ammunition products.This prediction can offer a decision basis for the inventory management of ammunition.In this paper,the evaluating parameters,the influencing factors of failure,types of storage life test and the data characteristic of ammunition are analyzed based on the features of ammunition storage reliability.A predicting algorithm on the data for normal stress storage life test is provided by combining an improved global particle swarm optimization algorithm and the BP neural network algorithm.For the data of accelerated life test,we establish a hybrid model with a statistical model and artificial intelligence algorithm to forecast the storage reliability.An evaluation and prediction system of the ammunition storage reliability is designed based on the database management theory and the BP neural network algorithm optimized by PSO.The paper is divided into four parts:(1)For the assessment of storage reliability of ammunition,the storage reliability and storage life of ammunition are provided as evaluation indexes.We analyze the factors affecting the storage failure of ammunition,especially the influence of temperature and humidity stress on the ammunition function.Then we determine the prediction indexs of ammunition storage reliability including sample size,storage temperature,storage humidity,storage time and storage failure number.A method based on Bayes theory is proposed to preprocess particular data of ammunition,such asdisordered data and zero failure data.(2)For the data of the storage life test under normal stress condition,the method of adding man-made noise is enlarged the small sample size of data.A combined forecasting model is provided by combining an improved global particle swarm optimization algorithm and the BP neural network algorithm(named as IGPSO-BP network).In order to make up for the slow convergence speed and easy to fall into the local minimum value of BP network,we add momentum and variable learning rate parameters,and introduce particle global search strategy to improve inertia weight and search area of particles.Experimental results show that,the IGPSO-BP network can improve the convergence speed and prediction accuracy of the algorithm.Compared with the BP neural network,PSO-BP neural network and prediction model traditional based on Bayes theory,the IGPSO-BP has stronger approximation ability.(3)For accelerated life test data about constant humidity step temperature,the sample size of ammunition,storage temperature stress,detection time and the number of corresponding failure are viewed as prediction indexes.The environmental factor method based on Bayes theory is quoted to deal with the storage time of step temperature.We establish an IGPSO-BP neural network prediction model and an acceleration mechanism prediction model to predict storage reliability.In order to verify the rationality of the model,a case study is implemented.The experimental results show that the IGPSO-BP network model has a great advantage in the prediction accuracy and calculation process,and verifies the validity and accuracy of the network model.(4)The ammunition storage reliability system is designed based on C#programming language in the Visual Studio 2012.It realizes the function of evaluation and prediction on ammunition storage reliability.