Research On On-line Speed Prediction Based On Cloud Platform For Multi-driving Cycles

Automobile networking refers to the realization of automobile information sharing with the outside world relying on modern communication and network technology.In recent years,with the rapid development of 5G communication and internet technology,automotive networking has gradually become a hot spot.And it is also an important development technology route of intelligent networked vehicles.Transferring part of complex computing tasks to the cloud is a reliable solution to solve the problem of the insufficient computing power of the current automatic vehicle processor.This paper mainly discussed the three key problems in the process of online speed prediction.And the main contents of this paper are as follows:(1)A method to classify speed cases according to the changing trend of the historical speed curve is proposed for the problem of the driving cycle clustering.And multiple methods for time-series similarity measurement are analyzed to compare the similarities and differences,especially Euclidian Distance and DTW(dynamic time warping)distance similarity measurement methods.Next,DTW distance is introduced into the clustering algorithm,and a k-means model based on shape clustering is established.Then,the influencing factors of the clustering model are discussed.On this basis,the influence of the input historical speed sequence length and the number of clusters on the clustering results of the above three models is explored.The results show that when a small value of the clustering number is assigned,the clustering results of each cluster model are more reasonable.And the K-means model based on shape clustering,which uses DTW distance measurement,has better classification performance.(2)To solve the problem of velocity series prediction,the convolution process of Conv1D(one-dimensional convolution neural network)is studied and introduced into the field of velocity prediction.Combined with the advantages of Conv1 D which can capture local features and the advantages of LSTM(long short-term memory neural network)which can learn the timing information,the Conv1D-LSTM speed prediction model and multi-step speed prediction variant model are proposed respectively.Next taking the Conv1D-LSTM model as the research object,the influence of prediction speed sequence length on its prediction performance is discussed.Then,based on the optimal clustering model,a speed prediction model for multi-driving cycles is established.And the performance of this model is compared with other models.The results show that the classification of speed plays a positive role in improving the performance of the prediction model,and the speed prediction model for multi-driving cycles in this paper has the best performance.(3)To solve the problem of online speed prediction,the Rocket MQ micro message center is designed based on the MQTT protocol.Based on these,a complete cloud platform which can provide real-time speed prediction service is established.Next,in order to improve the user experience and ensure the security of data transmission,the user single login module is completed based on the technology of JWT(JSON web token)and SSL(secure sockets layer)encryption technology.Then,the speed prediction model is dumped based on the ONNX framework technology,and the general model is embedded in the application service of the cloud platform.Functional test results show that the whole speed prediction service takes about 130 ms from message publishing to receiving the prediction results,which indicates that the architecture design of the cloud platform is reasonable and efficient,and has certain practicality.