Research On Identification Method Of Heavy Locomotive Adhesion State Based On Neural Network

The effect of the traction and braking performance of heavy-duty locomotives depends on the condition of wheel-rail adhesion.Adhesion generally includes four types: normal operation,slip symptoms,minor slip,and severe slip.The transition from slipping to micro slipping is very rapid.The development from micro slipping to severe slipping can cause wheel slipping and idle and derailment accident.At present,the research on the adhesion status identification of heavy-duty locomotives is mainly focused on the model-based approach.Because the physical structure of the heavy-duty locomotives is complex between wheels and rails,the interleaving effects between various devices are difficult to analyze thoroughly,so they cannot be modeled.Thorough mechanism analysis.In view of the above-mentioned deficiencies,according to the characteristics of data changes in the adhesion state of heavy-duty locomotives,this paper studies the methods of neural network,extreme learning and deep neural network based on genetic algorithm optimization.(1)The adhesion characteristics of heavy-duty locomotives were analyzed,and the classification of four adhesion states was defined based on the characteristics of heavy-duty locomotives' adhesion state from normal to skidding signs,minor skidding to severe skid transition.(2)Aiming at the current situation that the accuracy of the adhesion state recognition is not high,a neural network adhesion state recognition method based on genetic algorithm is designed.The method performs well on small scale test sets for the identification of adhesion state.(3)For the rapid change of engineering characteristics between the adhesion state,at the same time for the comparative analysis to find out more suitable for heavy-duty locomotive adhesion state identification method.Further,the adhesion state recognition method combining extreme learning machine theory with heavy vehicle locomotive adhesion characteristics was studied,and a data-driven adhesion state identification model was established to verify the feasibility of the proposed method.(4)For the problem of excavating a large number of samples with adhesive properties,this paper further studies the method of identifying the adhesion state of heavy-duty locomotives combined with deep neural networks.Simulation results show that this method performs well under large-scale samples.