Study on sensor data processing method based on generative neural network

The method of using experimental data as a training dataset to train deep neural networks for sensor measurement error compensation has aroused widespread research interest among researchers. Due to the difficulty and high cost of experiments, the sensor measurement data obtained through experiments is limited. Therefore, it is difficult to achieve the compensating method for sensor measurement errors by training deep neural networks with sufficient experimental data. In the case of limited sensor experimental data, a generative deep neural network is used in this paper to simulate the obtained limited amount of experimental data and generate a sufficient amount of experimental data. A mixed dataset is constructed using real experimental data and generated experimental data to complete the training of the error compensation network and achieve the correction of sensor measurement results. By comparing the sensor results corrected by measurement error correction model trained by original experimental data and the mixed data, it can be proven that the mixed data trained measurement error correction model has a better compensation effect on measurement errors.