Modeling anti-submarine warfare with machine learning and reinforcement learning algorithms

Submarines have been platforms that can directly affect the course of war from the moment they first appeared on the battlefield. While these platforms can be a great force multiplier when used as a friendly element, they will be a great risk when they are used as a threat. For this reason, it is very important to have an effective defense structure against platforms that have such a great impact in the field of operations. In this study, the necessary platforms (warship, aircraft and helicopter) and the number of weapons to be fired from these platforms were tried to be calculated in order to have an effective defense structure against threat submarines. In this context, firstly, the data to be used in the study were defined and the said data were pre-processed. In this regard, the parameters affecting submarine defense warfare were determined and a 220-line source data set was created from data with appropriate distributions for these parameters. Then, new data sets were created by using 8 different synthetic data generation techniques (the number of rows in the data set was increased to 10000 with each technique) to increase the number of data in this data set. Subsequently, a series of tests were applied to the created data sets to determine which technique would be used to increase data, and the synthetic data generation technique that gave the best results was determined. As a result of the examination, it was determined that the most suitable data augmentation method was the MDO method. Then, the number of weapons required to destroy each threat submarine was determined by using 10000 data created by the MDO method and 7 different machine learning models. In the next part of the study; First, threat submarine detection probabilities were calculated. Then, using detection probabilities, number of weapons to be fired and platform/weapon costs, and using the Reinforcement Learning algorithm, the results obtained from various operational modes were calculated to destroy all threat submarines. According to these calculations, the optimal mode of operation was determined. In the last part of the study, the gains to be achieved are mentioned.