Abstract

The phenomenon of decay，browning and decay is easy to appear during the table grape transportation，which causes no little financial loss.Cold－chain transportation is an effective measure to prevent those problems.At present，the cold－chain transportation construction is still in initial stage in china.It is urgent to conduct a research on the intelligent monitoring technology of coldchain transportation to complete data collection and processing.The way guarantees the quality of the table grapes and promotes the rapid development of table grapes cold－chain transportation.

In the paper，red globe grape was chosen as the research object.Wireless sensor network，multi－sensor data fusion technology and neural network theory were used for data acquisition and processing method combined with the preservation technology to guarantee the security of table grapes cold－chain transportation.By means of investigation research，emulation and simulation，some methods were put forward.Main conclusions were as follows:

(1) Temperature，relative humidity，and volume concentration of sulfur dioxide are the main factors influencing the quality of table grapes transportation，and the three factors were chosen as the the main monitoring parameters in the cold－chain transportation.Spatial difference of temperature and operation nonstandard were the main reasons for those factors changes.

(2) Multi－objective fuzzy matter element analysis method was put forward to optimize the sensor quantity.27sensors was reduced to 7 in the refrigerator car，which reduced the cold－chain transport costs.The statistical analysis method and the temperature field analysis were applied to validate the rationality of the optimization algorithm.Results possessed more than 95% of confidence level.

(3) A new data fusion method based on time－space data fusion theory was investigated to multi－sensor data processing.The mentioned algorithm could gradually reduce the influence of sensors with poor precision by means of introducing correction factor into the weighting coefficient，which taked advantage of the variance of the single－sensor fusion data and the final fusion data to adaptively adjust the weights of each sensor，and by means of multi－step fusion，gradually weakened the influence of some sensors with larger errors on fusion accuracy.Experimental results showed that the proposed algorithm outperforms traditional method.

(4) MGM－RBF neural network predictive model was put forward to predictive the refrigerator compartment temperature.Results showed that mean square relative error was 0.60% and average relative error was 0.44%，that were superior to single MGM arithmetic and RBF neural network prediction arithmetic.Based on the theory of statistical process control，three early－warning modes were defined.Compared with the fixed threshold method，the proposed algorithm had a less false－alarm rate.

(5) Monitoring research involved four dimensions: key parameters identification，parameters acquisition，parameters estimation and parameters forecast.Those formed an efficient and practical method for cold－chain transportation dynamic monitoring.

Keywords: Monitoring; Table Grapes; Cold－chain Transpotation; Data Fusion; Neural Network