基于BP神经网络的交流电源线连接器燃烧风险评估模型研究

A combustion risk assessment model of AC power cord connector based on the BP neural network

建立了一种基于BP神经网络的交流电源线连接器燃烧风险评估模型。通过环境试验和导体电化学迁移模型进行失效机理分析。失效分析表明,连接器样品在环境试验期间持续生成磷基酸,其能够对导体材料和绝缘材料产生腐蚀,存在引发短路的风险。在此基础上运用风险传递理论,结合火三角模型建立了风险传递路径。从人、物和环境3个角度阐释了影响因子的作用机理,构建了影响因子体系,在此基础上选用BP神经网络理论建立了交流电源线连接器燃烧风险评估模型。对178组样本数据进行归一化处理,随机抽取134组数据作为训练样本,剩下的44组作为检验样本,用于检验模型的正确率。训练检验共80次,模型的次均偏差为0. 2组,误差精度达到0. 004 5,模型结果满足精度要求。

Based on the BP neural network method,this paper intends to establish a model for combustion risk assessment in AC power cord connector in hoping to identify and determine the risk level of AC power cord connector. For this purpose,we have conducted a 28-day long test in the high-temperature and highhumidity environment on the AC power cord connector samples.The results of our test indicate that the mixed red phosphorus in the connector can be taken to serve as oxidizer,and the phosphorus acid in the oxidation product can be used as oxidizer to increase the environmental test effects. And,then,it would be possible to build up and establish an electrochemical migration model of the conductor from the microscopic point of view to analyze the reaction mechanism of the red phosphorus and the conductor material in the humid environment. In some cases,it can corrode the conductor material and the insulating material to get rid of the risk of short circuits. In the paper,we have proposed the failure mechanism analysis and the fire triangle model so as to build up and establish a risk transfer path. And,through the above mentioned failure mechanism analysis,it can be found that the mechanism of the action of the risk influential factors can be analyzed from the following 3 aspects: the material factors (products),the human operating factors and the environmental factors on the basis of the risk transfer path. And the influential system of the factors can be constructed and established based on the use of the BP neural network theory through the AC power cord connector combustion risk assessment model by comparing and analyzing the commonly used modeling methods. And,then,it would be necessary to train the sample data thoroughly to verify the correct rate of the model. The sample data can then be gained from the defective product administrative center of the state administration of the market regulation,which may have a total of 178 groups under its control and 178 sets of sampling data to be normalized. To avoid being over-fit,the paper has chosen 134 sets of data randomly as the training samples,whereas the remaining 44 sets were chosen to test the correct rate of the model as testing samples. And,then,through the above said careful training and testing,it would be possible to make the BP neural network risk assessment model of our results in an error accuracy of 0. 004 5 merely. Therefore,the model we have proposed tends to be qualified enough to be taken as a new research approach to the risk evaluation of such AC power cord connector.