A novel optimization method for WSN based on mixed matrix decomposition of NMF and 2-SVD-QR

Abstract WSN(Wireless Sensor Network), as the front-end of data acquisition of IoT(Internet of Things), its data always has the properties of massive, dynamic, correlated, and polymorphic, causing big troubles for the subsequent series of signal processing. Dense sensor nodes are randomly deployed in the monitoring field, so their perceived information cannot be independent so that there must be data redundancy. Therefore, considering the properties of complicated correlation in data acquisition, we propose a mixed matrix decomposition approach based on NMF(Non-negative Matrix Factorization) and 2-SVD-QR(Double-Singular-Value-QR) to optimize WSN: 1.Turn off redundant sensor nodes and retain a few nodes collectively to approximate the raw data output of WSN; 2.Explicit this method to largely eliminate the collected redundant data by the sensor in a coherent time. This approach not only reduces the amount of data that needs to be collected, but also significantly saves energy consumption and prolongs network lifetime. Experimental results verify that this method can effectively eliminate the correlation between the raw collected sensor data and highly improve the data (CR)compression ratio under the premise of ensuring the data reconstruction accuracy, and it is also better than existed WSN data compression approaches in terms of CR and reconstruction accuracy, which demonstrates the effectiveness of this mixed matrix decomposition.