Research On Short-term Traffic Flow Forecasting Of Wavelet Neural Network Based On Spark Distributed Ant Lion Algorithm

One of the conditions for the efficient operation of road traffic is real-time and ac-curate traffic flow prediction,which can not only help relevant transportation depart-ments to restrain and guide outgoing vehicles,but also one of the many topics worth studying in Intelligent Transportation Systems(ITS).In recent years,the accuracy of data-oriented intelligent modules has been continuously improved.One of the important reasons is that the more convenient and faster the acquisition of related road traffic de-tection data.Among them,the neural network structure in deep learning is widely used in various aspects such as road traffic sequence estimation,mainly due to the network imitating the learning method of the brain unit in the human body,and having the effi-cient ability of information processing and learning.However,reasonable plans and programs also need to be applied to how to make full use of complex massive data.In addition to managing and storing different types of data,comprehensive data can provide a rich training data set for complex models,but the data itself also makes project personnel face the challenges of large-scale sample processing and complex feature engineering.Therefore,the integration of massive data processing technology with the transportation industry has shown extraordinary depth and economic value.Using massive processing technology and Spark distributed multi-point computing,how to learn the data processing model that meets the expected requirements in a rea-sonable time has become the research topic of this article,and then try to create more complex application models.After analyzing and studying the mainstream multi-node remote computing framework Spark,this thesis proposes a parallel improved adaptive mutation elite weighted adjustment ant lion optimization algorithm based on the distrib-uted memory computing framework Spark.Its distributed design uses data parallel and computing parallel Hybrid optimization strategy,and applied to short-term road traffic estimation.The specific work is as follows:First,identify the traffic flow data from the California California Transportation Performance Measurement System(Peformance Measurement Sys-tem,PeMS),com-plete the pre-processing of data filling and hidden noise removal,and use the character-istics of sequence data to imply chaos to reconstruct High-dimensional complex con-struction of one-dimensional data to extract more hidden features.In order to accelerate the convergence speed of model learning,the training set is also normalized.After ana-lyzing the advantages and disadvantages of each model,a short-term road traffic estima-tion model was established on the basis of wavelet neural network(Wavelet Neural NetWork,WNN),and a model training experiment was conducted.Experiments prove that although the short-term road traffic estimation based on WNN is more in line with the actual trend,it still needs to be improved in terms of accuracy and stability.Next,the randomly generated initial values of the parameters to be optimized will affect the estimation effect of the wavelet neural network adjusted by the gradient cal-culation.This thesis introduces a new group optimization algorithm Ant Lion Optimizer(ALO)to improve the wavelet neural network parameter settings,design a new algo-rithm ALO-WNN for short-term road traffic estimation,and integrate it with the tradi-tional WNN network,and based on GA-WNN and PSO-WNN-based models were compared.The results show that although the prediction accuracy based on the ALO-WNN model has been improved to some extent,there is still room for improve-ment.In order to further improve the accuracy of ALO search,this thesis proposes an improved adaptive weighted elitism Ant Lion Optimizer(IWALO),which can be ob-tained through simulation experiment analysis.The IWALO-WNN model has high ac-curacy and overall performance In terms of aspects,they are better than the estimated effects of the models mentioned above.Finally,the typical common group optimization algorithm can only complete paral-lel iterations in a sequential mode simulation in a limited processor environment.This method is no longer suitable for large-scale growth of data sets in the era of big data.In order to solve the traditional algorithm model calculation The problem is that the vol-ume is large and the model design is complex,and large-scale training data cannot be effectively used.This thesis combines the IWALO-WNN algorithm model with the Spark distributed computing platform,and proposes a distributed design algorithm of data parallel and computing parallel fusion,which constitutes an improved adaptive mutation elite weighted adjustment based on Spark.Ant lion algorithm optimized wavelet neural network(Spark-IWALO-WNN)short-term road traffic estimation mod-el.