| 摘要 |
1. A cable shape prediction system comprising: a neural network comprising an input layer, a hidden layer, and an output layer, each layer comprising one or more nodes, nodes in the input layer being connected to operational data, at least one node in the input layer being connected to at least one node in the hidden layer and at least one node in the hidden layer being connected to at least one node in the output layer, the output layer outputting a predicted cable position, each connection between nodes having an associated weight; and a training apparatus for determining the weight for each said connection between nodes of the neural network, the neural network being responsive to the operational inputs for outputting a predicted cable position. 2. The system of claim 1 wherein the training apparatus comprises: apparatus for applying a plurality of training sets to the neural network, each training set consisting of historical data and a desired forecast cable position; apparatus for determining for each set of training data a difference between the forecast produced by the neural network and the desired forecast cable position; and apparatus for adjusting each weight of the neural network based on the difference. 3. The system in claim 2 wherein the training apparatus comprises apparatus for adjusting each weight by use of back propagation. 4. The system in claim 3 wherein the training apparatus further comprises means for applying a test data set to the neural network to determine whether training is complete. 5. The system in claim 4 wherein the test data set is a validation data set. 6. The system in claim 1 further comprising a preprocessor for computing a logarithmic value for each historical datum and for connecting each logarithmic value to the input layer. 7. The system in claim 1 wherein the neural network includes a bias node that has connections to at least one node in the hidden layer and at least one node in the output layer. 8. The system of claim 1 further comprising a normalizing apparatus for normalizing inputs to the neural network. 9. The system of claim 1 further comprising: at least one input for receiving at least one of vessel coordinates, receiver coordinates, time, vessel velocity, current velocity, wind velocity, water temperature, salinity, tidal information, water depth, streamer density and streamer dimensions as input to the neural network; and at least one output for generating a predicted cable shape output. 10. The system of claim 1 wherein the learning apparatus uses reinforcement learning. 11. A method for cable shape prediction comprising: providing operational data to a neural network input layer; and outputting a predicted cable position. 12. The method of claim 11 further comprising: applying a plurality of training sets to the neural network, each training set consisting of historical data, an associated statistical forecast cable shape made by the neural net and a desired forecast cable shape; determining for each set of training data a difference between the forecast produced by the neural network and the desired forecast cable shape; and adjusting each weight of the neural network based on the difference. 13. The method of claim 12 further comprising: adjusting each weight is performed by use of back propagation. 14. The method of claim 12 further comprising: applying a test data set to the neural network. 15. The method of claim 12 further comprising: applying a verification data set to the neural network. 16. The method of claim 11 further comprising: computing a logarithmic value for each historical datum and for connecting each logarithmic value to the input layer. 17. The method of claim 11 further comprising: applying a bias signal to nodes in the hidden layer and to nodes in the output layer. 18. The method of claim 11 further comprising: normalizing inputs to the neural network. 19. The method of claim 11 further comprising: receiving operational data as input to the neural network; and generating a predicted cable shape output. 20. The method of claim 11 further comprising: using reinforcement learning to train the neural network.
|
