Key Technology of Real-Time Road Navigation Method Based on Intelligent Data Research

Table 12

Comparison on characteristics of algorithms.

Algorithm	Characteristics
Kalman Filtering Algorithm	The model’s structure is relatively flexible and its solution process is simple and easy to implement. It is strongly adaptive. However, the initial values of the model are hard to be determined, which can influence forecasting effects. How to find a more suitable initial values is a direction for improvement of Kalman Filtering Algorithm
Neural Network Algorithm	The model is accurate in forecasting and steady as well. But its training speed is relatively slow. Besides, the setting of initial values that are connected to weight values in the neural network can directly influence speed of neural network training and eventual forecasting accuracy. In addition, the model cannot be popularized. Every roadway adopts its own model, so the model management is not very convenient. Thus how to improve the speed of network training is a direction of improvement of Neural Network Algorithm and also a demand of real-time traffic flow forecasting
Traffic Flow Forecasting Based on Time Series	The algorithm can carry out real-time forecasting. However, the model has many shortages. For example, the model has strict requirements for training data sets. Or its forecasting accuracy can be influenced greatly. Researches on time series provide many ways to improve the forecasting efficiency of the algorithm
Traffic Flow Forecasting Based on Transition Probability	The model is not complicated in computation and is able to quickly obtain model parameters. In addition, its forecasting accuracy is similar to the forecasting accuracy of Neural Network Algorithm; thus the model is suitable for forecasting real-time traffic flows. Meanwhile, Traffic Flow Forecasting Based on Transition Probability is easy to update and does not strictly require data to be in time series. Besides, the method is highly adaptive and flexible. However, the modeling and solution of the method are relatively complicated. So how to improve implementation of modeling and construction of training sets is a further direction for improvement