Research On Parking Guidance Technology In Parking Lot Real-time Data Missing Scenario

In recent years,the number of vehicles in my country has grown rapidly,and the problem of parking has become more prominent.The parking lots generally lack real-time data,which affected the construction of parking lots guidance system.Aiming at this problem,this thesis decomposes the problem of city-wide parking guidance and conducts research from the following two parts.Firstly,in light of the regional characteristics of parking demand,we designed a parking lots subnet segmentation model.The model uses the Mean Shift algorithm to segment the parking lots,which is based on the parking lots’ spatial relationships and initial influence.For the parking lots in a subnet,we calculated the effective influence of the parking lots based on the traffic dynamic model,which derived from the Page Rank algorithm,and used an improved weighted Voronoi diagram featuring boundary constraint to map the ranking value of each parking lots to the geospatial space.In this way,we realized the division of the parking lots influence and reduced the complexity of the parking lot network.Secondly,for the problem of parking demand in the parking lots subnet,we proposed an idea of parking guidance system by the crowdsourced parking data.The model is different from the traditional parking guidance system which based on the real-time data.We extracted the parking events by the time,trajectory and speed of the historical parking data.At the same time,the regional traffic condition were calculated based on the social travel trend in the same period.The two types of data are used as the input of the conditional Long Short-Term Memory(LSTM)deep learning model.The model was optimized from the perspective of classification.Trained by the crowdsourced data set,vehicles with clear parking needs can be classified into corresponding parking lots,and the parking guidance system can work effectively.The experiments were simulated on the actual road network in Shenzhen,China.We evaluated the effectiveness of the above two algorithms respectively.In the experiment,1,096 parking lots in Luohu District were divided into different parking subnets,and their effective influence coverage map is drawn.And also,an experimental area containing 14 parking lots was constructed for vehicles going to Shenzhen Children’s Hospital to test the effectiveness of the parking guidance system.The experiment shows that the model in this thesis has similar capability to traditional utility models when the traffic is unblocked,and the parking guidance accuracy reaches 97.75%.Compared with the model without regional traffic volume,the model guidance accuracy rate of this thesis is improved by 6.7%.And the model performs better under extreme traffic conditions.The experiments verify the effectiveness of our method for parking guidance without real-time parking data.