Research On Situation Assessment Technology Of Multi-source Data Based On Machine Learning

With the rapid development of technology,modern combat forms have become more complex and changeable.Facing the massive battlefield multi-source situation information obtained from the combat environment,it is quite difficult for the commander to make real-time decisions.The data sources are complex and diverse,with ambiguity and uncertainty,and the amount of data is extremely large.How to eliminate irrelevant information and dig out valuable feature information from these massive and complex information,so as to accurately reflect the current battlefield situation changes in real time to assist commanders to make correct operational decisions has become a hot issue in current research.The situation assessment technology of Multi-source data can fuse the battlefield multi-source situation information,and finally establish the current battlefield situation assessment map to help the commander make decisions.However,in the face of the complexity of the modern combat environment,situation assessment technology of Multi-source data needs to be improved to help the commander clarify the battlefield situation and make a rational analysis,so as to formulate a reasonable plan for the next plan more quickly and scientifically.Based on the functional model of situation assessment,this thesis has conducted in-depth research on the technical issues of Multi-source data situation assessment and the construction of Multi-source data situation assessment system.It mainly includes algorithm models in two key technologies about target grouping and target intention recognition in situation assessment,and design and implement a situation assessment system of Multi-source data.The research work content and results are as follows:1.The partition clustering algorithm in target clustering is studied and analyzed.Aiming at the sensitive problem of the initial clustering center in the K-Means algorithm,the sorted-D~2 sampling method is proposed,and on this basis,the SK-Means++algorithm is proposed.This algorithm effectively reduces the possibility that the initial clustering center is selected in the same cluster,and alleviates the problem that the partitioning clustering algorithm tends to fall into the local optimal solution.Compared with the K-Means++and K-MeansII algorithms,experiments prove that the SK-Means++algorithm has the best clustering effect.Aiming at the instability of clustering results of K-Means++algorithm,the double-D~2 sampling method are proposed.Using this method,a new partition clustering algorithm,K-MD~2 algorithm is proposed.At approximately the same time complexity,the clustering accuracy and stability are better than the K-Means++algorithm.Aiming at the problem that the traditional clustering algorithm has a high time complexity and cannot meet the real-time requirement in large data sets,a fast and efficient K-MHMD algorithm is proposed,and its time complexity is approximately sub-linear.Absolute advantage in large data sets.Compared with the K-MC~2 algorithm with the same sublinear time complexity,experiments prove that the clustering effect of the K-MHMD algorithm is better than the K-MC~2 algorithm.Finally,several groups of situation simulation data sets are used to compare different algorithms,the clustering effect of SK-Means++algorithm is better than other algorithms.2.The fuzzy clustering algorithm in target clustering is studied and analyzed.Aiming at the problem of fuzzy clustering algorithm using traditional distance division,a weighted Euclidean distance is proposed,and Pearson correlation coefficient is introduced to dynamically assign feature weight coefficients.Using this method,a self-adaptation possibility fuzzy c-means algorithm is proposed.The algorithm avoids the large impact of the less important features due to the consistency of all feature weights on the clustering result,so that it can quickly converge,improve the accuracy of clustering,and the algorithm can effectively reduce the probability of division errors due to the closer distance between different target clusters in air combat.Compared with the two fuzzy clustering algorithms mentioned in the article,the SA-PFCM++algorithm is the most stable and the clustering accuracy is higher.3.The target intention recognition algorithm is studied and analyzed.Combining the residual mechanism and the attention mechanism,an attention-based residual long short-term memory network model is proposed.Aiming at the problem of long-term information loss in the LSTM network,the model can capture the important information of the correlation between the upper and lower levels of the current time step and between the previous and subsequent time steps.Experiments show that the recognition ability of this model is better than the three intention recognition models mentioned in the article.4.The situation assessment system of Multi-source data is designed and implemented.In the system,the algorithm part of the target grouping module uses SK-Means++and SA-PFCM++algorithms,and the algorithm part of the target intent recognition module uses the ResLSTM model,and the design of each module and the storage method of situation data are explained in detail.This thesis develops a complete Multi-source data situation assessment system with browser as the core and Django as the framework.Compared with the previous situation assessment system,the system platform provides a more friendly human-computer interface,rich and colorful visual display,simple operation,and easy to master.