| Traffic signal control is one of the important measures to guarantee the safety of traffic flow and alleviate urban traffic congestion.It can be divided into vehicle actuated control,adaptive control and fixed time strategy.According to the dynamic optimization control scheme of real-time traffic flow data,the vehicle actuated control and adaptive control can well adapt to the short-term random fluctuation of traffic demand.Fixed time strategy,also known as Time of Day(TOD),divides a time unit(usually one day)into several time periods according to the changing of traffic demand,and then optimize the signal control scheme in each time period.This kind of method has a good control effect in the low and medium saturation traffic flow.For any signal control system,the fixed time strategy is usually regarded as the default scheme in case of detector damage and data transmission failure.Fixed time strategy method includes time series partitioning and signal scheme op-timization.At present,there are many researches on the signal scheme optimization.The traditional methods for time series partitioning is mostly based on historical data and is accomplished by clustering method.There are two major defects:one is that the time char-acteristic of traffic flow scries are ignored,and the results cannot always be the global optimization of time series partitioning;The other is that this method cannot do well in forecasting of the traffic flow.When there is a big difference between the real data and the historical data,the partitioning scheme and the actual data will be mismatched.In order to remedy the above shortcomings,this paper carries out a long-term traffic flow forecasting and time series partitioning method for time-of-day breakpoints.At present,there are a few studies on long-term traffic flow forecasting.The existing methods classify the historical data by visual factors(e.g.day in week)mostly,and then the matched historical data can be selected out.Finally,the statistical analysis method can be used to forecast.However,the traffic flow parameters is the result of the superposition of multiple factors,including weekdays,weather,seasons,holidays and so on.Firstly,this paper analyses the periodic characteristics of traffic flow,then constructs a pattern cluster-ing method based on density peak clustering,a pattern matching method and a long-term traffic flow prediction based on neural networkSecondly,the time series partitioning method based on K-means clustering idea has many defects,so this paper proposes two improved algorithms based on the optimization of K-means.Then,considering the time order,the optimal switching of traffic flow parameters can be realized by time series partitioning,and a dynamic recursion method can be used for optimization.Finally,our method of the practical effect is verified by simulation test with the actual traffic flow data of several cities.These results show that the forecasting method can sig-nificantly improve the prediction accuracy of traffic flow parameters,and the time series partitioning method can effectively improve the overall operation efficiency of traffic flow. |
