Prediction Of Driving Condition For Plug-in Hybrid Electric Vehicles

The speed prediction of vehicles,as one of the important research subfields ofenergy saving and security, can be used for auxiliary driving of energy saving andsafety, route guidance, gears control of AT (automatic transmission),and the predictivecontrol of HEVs(hybrid electric vehicles).So the speed prediction has importanttheoretical basis and practical value. The major contributions of the thesis are asfollows:①Base on analysis of variance of single factor and correlation analysis, five inputcharacteristic parameters of BP neural network speed prediction model are determinedas below: average speed, ratio of idling time, value of speed multiplying accelerationvariance, velocity variance and average value of positive accelerations.By thedistribution of boxplots and statistical data of the above five characteristic parameters,6kinds of conditions of the speed prediction are identified by six classificationboundaries of six BP neural network speed prediction submodels, which all regardaverage speed and ratio of idling time as thresholds.②Based on the theory of BP neural network, two speed prediction model of BPneural network with8input nodes and60output nodes are established respectively.Each speed prediction model includes six submodels,the inputs of which all include theabove five characteristic parameters and historical neighboring speed. According to thetraining and prediction results of two models, statistical analysis of vehicle workingconditions, the length of input historical neighboring speed and the number of outputnode are modified, then speed prediction model of BP neural network with revisedinput/output node is established, which achieves better effects on the training andprediction than before.③According to distribution differentiation of weight threshold, differentiatedinitialization of initial population of genetic algorithm is applied, weight threshold ofBP neural network speed prediction model are optimized by the genetic algorithm. Theoptimized BP neural network speed prediction model has a better effects on the trainingand prediction than before.④According to the change of fitness values of particle swarm algorithm, anadaptive adjustment formula for inertia weight is proposed to improve the algorithm.Ahead of the optimization of the BP neural network speed prediction model by particle swarm algorithm, initial population of genetic algorithm is initialized based ondistribution differentiation of weight threshold, and weight threshold and search rangeof speed are set respectively in the process of optimization. The effects on the trainingand prediction of optimized BP neural network speed prediction model by particleswarm algorithm are improved compared prediction method is better than the other twomethods.⑤Based on the optimization results of BP neural network speed prediction modeloptimized by genetic algorithm and particle swarm algorithm, PSO-GA Jointoptimization algorithm model is put forward. Under the urban traffic conditions and thehighway (1,2,6speed prediction) conditions, BP neural network speed predictionsubmodel optimized by genetic algorithm is adopted. And under the urban unimpededtraffic conditions and suburb (3,4,5speed prediction) conditions, BP neural networkspeed prediction submodel optimized by particle swarm algorithm is utilized. Theeffects of BP neural network speed prediction model is verified better based on thetraining and prediction when optimized by PSO-GA Joint optimization algorithmother,superior to the use of GA and PSO algorithm separately.