Research On Prediction Of Off-base Height Of Marine Electromagnetic Transmitter Based On Deep Learning

Through years of continuous development,the marine controlled source electromagnetic method has been widely used by geophysical prospectors worldwide for the exploration of natural gas hydrates,submarine oil and gas resources.The effectiveness of the method has been fully verified.With the increasing application of marine controlled source electromagnetic method,some technical problems occur,which need to be resolved.For example,when the detection is performed,the height between the transmitter and the sea bottom has to be stable within a certain range,so that the submarine electromagnetic data with good consistency can be collected.In order to solve this problem,this work introduces a deep learning-based underwater high-power electromagnetic transmitter control scheme : using deep learning to make a nonlinear regression on the transmitter drag curve to predict the height of the transmitter at next moment,which will be used to guide the ship staff to make operations in advance to make real-time adjustments.Nowadays,deep learning,a typical category of artificial intelligence,has become a hot research area and has been applied in many practical fields,such as speech recognition and image processing.By designing multi-layer neural network,a large number of simple computing units are connected together to achieve the capability of expressing complex relationships between inputs and outputs.According to the error evaluation of ten times of ten-fold cross validation,this work selects the long shortterm memory(LSTM)as a main body to build the network.LSTM is a special kind of recurrent neural network(RNN),which introduces self-loop to make the gradient flow continuously,avoiding the gradient disappearance and gradient explosion during the long sequence training.In other words,compared to ordinary RNN,LSTM is able to perform better in the longer sequence.The data is preprocessed to obtain a training set containing 10,000 labeled samples.After parameter adjustments,the trained model has good performance on prediction(relative error is less than 5%),and has strong practicality,robustness and generalization ability.The entire model is implemented on python,using Pycharm as a language development software and Anaconda to set up language environment.In this thesis,deep learning is introduced in the marine controlled source electromagnetic detection.By using the data transmitted from the hardware,the future off-base height of transmitter can be predicted at any time of the towing process.After that,the operation instructions such as casting or retracting the cable with specific length and speed are given to the ship staff.It can effectively prevent the transmitter from hitting the bottom and ensure that the off-base height of the transmitter is stable within a certain range.The proposed method can achieve better accuracy and stability compared with previous manual observation and judgment,providing technical supports for obtaining high-quality submarine electromagnetic data.