Neural Network-based Power Line Channel Impedance Estimation Method

With the development of smart grids,Power Line Communication(PLC)technology has become an important technology for information exchange between users and devices in power transmission networks.In order to achieve high-quality power line communication,it is necessary to have a clearer understanding of the power line channel.Therefore,channel impedance prediction has become an important content in the field of power line communication,and a channel load impedance estimation method is proposed.First,in order to build a relatively real channel model,a variety of power line channel structure models were studied and analyzed.MATLAB simulation was used to build a large number of channel models of single-branch load structures with different load impedances and single-node multi-branch load structures,and a large number of different load impedances were obtained.System functions under.Secondly,I studied the Back Propagation(BP)algorithm in depth,and built a BP neural network with MATLAB.Use the obtained system function to introduce noise to generate learning samples,reduce the interference of noise on BP neural network training through the denoising algorithm,compare the prediction errors before and after denoising,and obtain a more accurate single-branch load The BP neural network for predicting the structure realizes the prediction of load impedance.Finally,the Field Programmable Gate Array(FPGA)power line communication simulation prototype was used as the transceiver terminal to construct the transfer function measurement platform of the single branch power line structure,and the transfer functions of different load impedances were obtained.Then through the IFFT transformation of the low-frequency stable signal,and the cross-correlation of the P symbols in the OFDM preamble sequence after the transformation to eliminate jitter and reduce the error,it is found that the impedance value of different loads can be determined by identifying the value range of the cross-correlation value.It is estimated that the prediction of load impedance can be realized again.Adopt tsoc FPGA development board DE10-nano,realize BP neural network based on OpenCL.By superimposing SNR=20dB noise.Comparing the training and test results of the original sample and adding 20 dB noise,it is found that under noise interference,the training time required to reach 100% of the recognition success rate is longer than that under no noise conditions.