TWI703517B - System for bus arrival time prediction and method thereof - Google Patents

Abstract

This invention provides a system for bus arrival time prediction and a method thereof, which uses GPS information of a vehicle and arrival/departure information to count the departure time of the vehicle at a first position in a travel route, such that the travel time of the vehicle for the next departure time from the first location to the second location of the travel route can be predicted according to the departure time, the national highway travel time of the vehicle of the departure time from the first position to the second position of the travel route of the national highway, and the historical travel time between the first location and the second location.

Description

System and method for predicting vehicle arrival time

The present invention relates to travel time prediction technology, especially a system and method for vehicle arrival time prediction.

With the development of cities and the increase in population, the traffic volume of the transportation system connecting the cities has increased sharply. However, the traffic travel time prediction of the traditional classification method is often due to the same time label (date and time label within the same week). The average value of the actual driving time data is used as the output value of the predicted time. However, the definition of the date and time label is often unclear, such as the Friday before the weekend and the Friday before the consecutive holiday, or the definition for each day of the consecutive holiday. It is difficult to define its attributes, and because of the lack of sample data under the same label in the past, the use of this statistical forecast and the actual results cause errors.

Therefore, how to effectively improve the error of the predicted travel time caused by the date and time label and predict the travel time more quickly and accurately is the technical problem to be solved by the present invention.

In order to solve the above problems, the present invention provides a vehicle arrival time prediction system, which includes: a vehicle-mounted mobile device, which is configured on the vehicle to obtain GPS information and arrival/departure information of the vehicle; and a server, which includes : The receiving module receives the driving route of the vehicle and the GPS information and arrival/departure information of the vehicle transmitted by the on-board mobile device. The driving route includes a first position and a second position; the statistical module is According to the GPS information and arrival/departure information, count the departure time at the first location of vehicles traveling on the driving route in successive shifts; the reading module receives the departure time of each vehicle from the first location to the first location The national road travel time of the second location on the national road; the historical database is to store the historical driving time between the first location and the second location of the driving route for each vehicle at the departure time in the history of the departure time/departure time Time; and the prediction module, which uses machine learning to predict the vehicle at the next departure time from the first position of the driving route based on the travel time of the national road and the historical travel time of each vehicle at the departure time Travel time to the second location.

In one embodiment, the statistics module calculates the departure time of the two consecutive shifts of the driving route at the first location based on the GPS information and arrival/departure information.

In one embodiment, the reading module reads the national road travel time of the national road traveled by the vehicle traveling from the first location to the second location from the national highway electronic toll collection system based on the GPS information.

In one embodiment, the machine learning system is LSTM.

In one embodiment, the structure of the LSTM includes: a normalization layer, which normalizes the national road travel time and the historical travel time of each vehicle at the departure time and each departure time; the conversion layer, which normalizes the normal The converted data is converted into the time-level data for training and learning of the LSTM; the LSTM model is used to input the time-level data with two time-levels and output one-time level The training and learning of the discarded layer, which sets the label retention rate of the training and learning to reduce the complexity of the LSTM model after the training and learning; and the fully connected layer, which outputs the reduced complexity of the LSTM model to make the The output first time level is the travel time of the vehicle from the first position to the second position of the driving route for the predicted next departure time.

The present invention also provides a method for predicting the arrival time of a vehicle, which includes: enabling the receiving module to receive the driving route of the vehicle and the GPS information and arrival/departure information transmitted by the vehicle-mounted mobile device on the vehicle, wherein the driving route Including the first location and the second location; make the statistics module calculate the departure time at the first location of the continuous shift of the driving route based on the GPS information and arrival/departure information; make the reading module read each The travel time of the national road traveled by the vehicle from the first location to the second location at the departure time; order the historical database to provide the vehicle with the departure time in the history of the departure time/departure time from the travel route The historical travel time between the first position and the second position; and make the prediction module use machine learning to predict the next departure based on the national road travel time and the historical travel time of each vehicle at the departure time and each departure time Time is the travel time of the vehicle from the first position to the second position of the driving route.

In one embodiment, the statistics module is allowed to count the departure time of the vehicles traveling on the driving route for two consecutive shifts at the first location based on the GPS information and arrival/departure information.

In one embodiment, the reading module is made to read the national road travel time of the national road traveled by the vehicle traveling from the first position to the second position from the national road electronic toll collection system based on the GPS information.

In one embodiment, the machine learning system is LSTM.

In an embodiment, the LSTM predicting the travel time of the vehicle at the next departure time from the first position to the second position of the driving route includes: using a normalization layer to divide the travel time of each vehicle The travel time of the national road and the historical travel time of each vehicle at the departure time are normalized; the normalized data is converted into the time-level data for the LSTM training and learning using the conversion layer; the LSTM model is used to input two The training and learning of the time-level data is carried out in the way of outputting the time-level data; the discarding layer is used to determine the label retention rate of the training and learning to reduce the complexity of the LSTM model after the training and learning; and the fully connected layer will reduce the The output of the LSTM model is complicated, so that the first time order of the output is to predict the travel time of the vehicle from the first position to the second position of the driving route at the next departure time.

It can be seen from the above that the present invention discards the time label part of the date and time label, and only uses the time period time label, and cooperates with LSTM to train and learn the time parameters under the time period time label to predict the travel time of the time period time label. In order to improve the error of the predicted travel time caused by traditional date and time tags, and to predict the travel time of vehicles with time tags (such as national highway passenger transport) during the time period of the time tag to accurately predict the vehicle (such as national highway passenger transport) The arrival time of each shift period will increase the people's willingness to use national highway transportation services.

1‧‧‧Car mobile device

2‧‧‧Server

21‧‧‧Receiving Module

22‧‧‧Statistics Module

23‧‧‧Read Module

24‧‧‧History Database

25‧‧‧Prediction Module

31‧‧‧Regularization layer

32‧‧‧Conversion layer

33‧‧‧LSTM model

34‧‧‧Discard layer

35‧‧‧Fully connected layer

S1~S5, S51~S55‧‧‧Step

Figure 1 is a schematic architecture diagram of the vehicle arrival time prediction system of the present invention; Figure 2 is a schematic architecture diagram of the LSTM of the present invention; Figure 3 is a step flow chart of the vehicle arrival time prediction method of the present invention ; And Figure 4 is a step flow chart of the step S5 of the present invention to use LSTM to predict the next departure time of the vehicle from the first position to the second position of the driving route.

The following specific examples illustrate the implementation of the present invention. Those familiar with the art can easily understand the other advantages and effects of the present invention from the contents disclosed in this specification.

It should be noted that the structures, proportions, sizes, etc. shown in the drawings in this manual are only used to match the contents disclosed in the manual for the understanding and reading of those familiar with the art, and are not intended to limit the implementation of the present invention Therefore, it does not have any technical significance. Any structural modification, proportional relationship change, or size adjustment should still fall within the scope of the present invention without affecting the effects and objectives that can be achieved. The technical content disclosed by the invention can be covered.

Figure 1 is a schematic architecture diagram of the system for predicting vehicle arrival time according to the present invention. In this embodiment, the vehicle may be, for example, passenger transportation. As shown in Figure 1, the system includes: a vehicle-mounted mobile device 1 and a server 2, wherein the vehicle-mounted mobile device 1 and the server 2 are connected to each other through a wireless network.

The vehicle-mounted mobile device 1 is configured on a vehicle (such as passenger transportation) to obtain GPS information of the vehicle, and compare the GPS information with the position information pre-stored in the vehicle-mounted mobile device 1 to generate arrival/departure Station information. In this embodiment, the location information is stop sign information for passenger transportation.

The server 2 includes: a receiving module 21, a statistics module 22, a reading module 23, a historical database 24, and a prediction module 25.

The receiving module 21 receives the driving route of the vehicle and the GPS information and arrival/departure information of the vehicle transmitted by the in-vehicle mobile device 1. The driving route includes a first location (such as a first stop) and a first stop. Second position (such as the second stop sign).

The statistics module 22 calculates the departure time/departure time of vehicles (such as passenger transport) in the first location (such as the first stop sign) of the continuous shift of the driving route based on the GPS information and arrival/departure information.

In one embodiment, the statistics module 22 counts the number of vehicles (such as passenger transport) on the first location (such as the first stop sign) of the two consecutive shifts traveling the route based on the GPS information and arrival/departure information. Departure time/departure time.

In one embodiment, the departure time of the vehicles of the consecutive shifts continuously has a time period difference, and the time period difference may be in units of hours or minutes, but is not limited to this.

The reading module 23 is used to receive the national highway that the vehicle (such as passenger transport) of each departure time/departure time travels from the first position (such as the first stop) to the second position (such as the second stop) National highway travel time.

In one embodiment, the reading module 23 reads the departure time/departure time of the vehicle (such as passenger transport) from the first location (such as the first stop sign) to the national highway electronic toll collection system based on the GPS information The national road travel time of the second location (such as the second stop sign).

The historical database 24 stores the departure time/departure time of the vehicle (such as passenger transport) from the first position (such as the first stop sign) and the second position of the driving route during the history of the departure time/departure time. For example, the historical travel time between the second stop sign), where the historical travel time includes historical national road travel time and historical non-national road travel time (ie, historical travel time of general roads).

In one embodiment, the historical date of the historical departure time slot/departure time is the same as the day of the departure time/departure time. For example, if the day of the departure time/departure time is Monday, then the historical departure time/ The historical date of the departure time is the Monday of a different week/week in the history, but the historical date may be different from the date, so it is not limited to this.

In one embodiment, the historical driving time in the historical database 24 is calculated by a calculation module (not shown) based on the GPS information and arrival/departure information of the vehicle received by the receiving module 21 And the historical driving time of the calculation result is stored in the historical database 24, but not limited to this.

The prediction module 25 uses machine learning to predict the next departure time slot/departure time based on the national road travel time and the historical travel time of the vehicle (such as passenger transportation) of each departure time slot/departure time. Time is the travel time of a vehicle (such as passenger transportation) from the first position (such as the first stop) to the second position (such as the second stop) of the driving route.

In one embodiment, the next departure time slot/departure time and each departure time slot/departure time successively have the time interval difference, for example: each departure time slot/departure time is 2, the time interval difference is 15 minutes, and the If the two departure time slots/departure time are 8:00 and 8:15, the next departure time slot/departure time is 8:30, but not limited to this.

The machine learning system performed by the prediction module 25 is LSTM (Long Short-Term Memory).

Figure 2 is a schematic structural diagram of the LSTM of the present invention. As shown in Figure 2, the LSTM system includes: a normalization layer 31, a conversion layer 32, an LSTM model 33, a discarding layer 34, and a fully connected layer 35.

The normalization layer 31 normalizes the national road travel time and the historical travel time of each of the departure time/departure time and each of the departure time/departure time of the vehicle (such as passenger transportation). In an embodiment, the normalization layer 31 allocates the normalized data between 0 and 1, but it is not limited to this.

The conversion layer 32 converts the normalized data into time-level data for training and learning of the LSTM.

The LSTM model 33 trains and learns the data of this time level by inputting two time levels and outputting one time level.

The discarding layer 34 sets the label retention rate of the training and learning to reduce the complexity of the LSTM model 33 after the training and learning.

The fully connected layer 35 outputs the reduced complexity of the LSTM model 33, so that the first order of the output is to predict the next departure time/departure time of the vehicle (such as passenger) from the first position (such as the first The travel time from the stop sign) to the second position (such as the second stop sign).

Figure 3 is a flow chart of the method for predicting vehicle arrival time according to the present invention. As shown in Figure 3, the method includes the following execution steps: In step S1, the driving route, GPS information, and arrival/departure information are received. The receiving module 21 is made to receive the driving route of the vehicle (such as passenger transportation) and the GPS information and arrival information transmitted by the vehicle-mounted mobile device 1 on the vehicle. / Departure information, where the driving route includes a first position (such as the first stop sign) and a second position (such as the second stop sign).

In one embodiment, the arrival/departure information is generated by the vehicle-mounted mobile device 1 based on the GPS information and the site information pre-stored in the vehicle-mounted mobile device 1.

In step S2, the departure time/departure time is counted, and the statistics module 22 is made to count the vehicles (such as passenger transport) traveling the driving route in the first position (such as passenger transportation) based on the GPS information and arrival/departure information. The first stop) departure time/departure time.

In one embodiment, the statistics module 22 is made to count the vehicles (such as passenger transportation) of two consecutive shifts traveling the route based on the GPS information and arrival/departure information at the first location (such as the first stop sign) Departure time/departure time.

In one embodiment, the departure time of the vehicles of the consecutive shifts continuously has a time period difference, and the time period difference may be in units of hours or minutes, but is not limited to this.

In step S3, the national road travel time is obtained, wherein the reading module 23 is made to read the vehicle (such as passenger transport) of the departure time slot/departure time from the first position (such as the first stop) to the second The national road travel time of the location (such as the second stop sign).

In one embodiment, the reading module 23 is made to read the departure time/departure time of the vehicle (such as passenger transport) from the first location (such as the first stop sign) from the national highway electronic toll collection system based on the GPS information The travel time of the national highway to the second location (such as the second stop sign).

In step S4, the historical travel time of the travel route is obtained, wherein the historical database 24 is asked to provide the vehicle (such as passenger transport) of the departure time period/departure time from the historical travel time period/departure time of the travel route. The historical travel time between the first location (such as the first stop) and the second location (such as the second stop).

In one embodiment, the historical date of the historical schedule/departure time is the same or different from the date of the historical schedule/departure time, but not limited to this.

In one embodiment, the historical driving time in the historical database 24 is determined by a calculation module (not shown) based on the GPS information and arrival/departure information of the vehicle received by the receiving module 21. The station information is calculated, and the historical driving time of the calculated result is stored in the historical database 24, but not limited to this.

In step S5, machine learning is used to predict the travel time of the next departure time slot/departure time of the vehicle (such as passenger transport), wherein the prediction module 25 is made to use machine learning to predict each departure time slot/departure time and each departure time slot/ The national road travel time of the vehicle (such as passenger) at the departure time and the historical travel time, and then the next departure time/departure time of the vehicle (such as passenger) from the first position (such as the first stop) of the driving route to The travel time of the second location (such as the second stop), where the machine learning department is LSTM.

In one embodiment, the next departure time slot/departure time and each departure time slot/departure time successively have the time interval difference, for example: each departure time slot/departure time is 2, the time interval difference is 15 minutes, and the If the two departure time slots/departure time are 8:00 and 8:15, the next departure time slot/departure time is 8:30, but not limited to this.

Figure 4 is the step S5 of the present invention using LSTM to predict the next departure time/departure time of the vehicle (such as passenger) from the first position (such as the first stop) to the second position (such as the second stop) of the driving route ) The flow chart of the steps of driving time. As shown in Figure 4, the LSTM system includes the following execution steps: In step S51, normalization, wherein the normalization layer 31 is used to normalize the national road travel time and the historical travel time of each of the departure time slot/departure time and each of the departure time slot/departure time of the vehicle (such as passenger transportation) . In an embodiment, the normalization layer 31 allocates the normalized data between 0 and 1, but it is not limited to this.

In step S52, the data is converted, wherein the normalized data is converted into the time-level data for training and learning of the LSTM by the conversion layer 32.

In step S53, the LSTM model is used for training and learning, and the LSTM model 33 is used to train and learn the time-level data by inputting two time levels and outputting one time level.

In step S54, the complexity of the LSTM model is reduced, wherein the discarding layer 34 is used to determine the label retention rate of the training and learning, so as to reduce the complexity of the LSTM model 33 after the training and learning.

In step S55, the result is output, wherein the fully connected layer 35 is used to output the reduced complexity of the LSTM model 33, so that the first order of the output is to predict the next departure time/departure time of the vehicle (such as passenger transport) from the The travel time from the first position of the route (such as the first stop) to the second position (such as the second stop).

The following provides a list of time series variables by way of example to illustrate step S5 of the method for predicting the arrival time of vehicles (such as passenger transport) of the present invention:

Assuming that one of the passenger transportation routes in Table 1 is from Taipei City Hall Transfer Station (Stop A) to Yilan Luodong Transfer Station (Stop B), the country's road line through which the traveling route passes is distinguished by the establishment of the National Highway Electronic Toll Collection System It is a road section of five countries, and the time-level data for the LSTM training learning is set to 15 minutes (as shown in Table 1 for each departure time/departure time with a time difference of 15 minutes), of which 15 minutes is only an example. Adjust the actual situation.

The actual input and output data of the LSTM model 33 are shown in Table 2. The output data is the predicted travel time y from stop A to stop B at 08:30 on 2019/07/01. The input data is the first two hours Time series data (variables X1~X14) of the order (2019/07/01 08:00 and 08:15), where the variables X1~X7 and X8~X14 are respectively represented as the departure time/departure time to this time level in Table 1 The value between the historical travel time from the departure to the target stop. In another embodiment, the input data can also be the first N time levels, and each of the first N time levels has 7 variables (as shown in Table 2).

According to the X1~X14 parameters in Table 2 above, after executing steps S51~S55, it can be predicted that the travel time y from stop A to stop B at 08:30 on July 1, 2019 is 83 minutes, and it is fully connected The LSTM model output by layer 35 is the prediction model model_predict for passenger arrival time prediction, where the normalization layer 31 in this step S51 compares time-level type data (such as X1 in Table 2) and travel time type data (such as X2~X7) in Table 2 have different normalization ratios.

The prediction model model_predict can be expressed as model_predict(data_real,in_step,out_step,n_feature,n_LSTM), and according to the variables in the prediction model model_predict shown in Table 3, the data collected in real time (as in this example above X1~X14 ) Perform the step S5 to predict the travel time from departure to the target stop during the time period of this shift or the time period of the next shift.

It can be seen from the above that the present invention uses only the time label (that is, the time period difference between each departure time/departure time) by discarding the time label part of the date and time label, and cooperates with the LSTM to compare the departure time with the time label of the time period. The time parameters under the time slot/departure time are trained and learned to predict the travel time of the time slot time label, thereby improving the error of the predicted travel time caused by the traditional date and time label, and carry out the flight time of the national highway passenger transport with the time slot time label The travel time prediction of the time label in this period can accurately predict the arrival time of the national highway passenger transport in each shift period, thereby increasing the people's willingness to use the national highway transportation service.

The above-mentioned embodiments only exemplify the principles and effects of the present invention, and are not used to limit the present invention. Anyone who is familiar with this skill can do so without departing from the spirit of the present invention and Under the scope, the above-mentioned embodiments are modified and changed. Therefore, the protection scope of the present invention should be as listed in the scope of patent application described later.

1‧‧‧Car mobile device

2‧‧‧Server

21‧‧‧Receiving Module

22‧‧‧Statistics Module

23‧‧‧Read Module

24‧‧‧History Database

25‧‧‧Prediction Module

Claims (10)

A system for predicting the arrival time of a vehicle includes: a vehicle-mounted mobile device, which is configured on the vehicle to obtain GPS information and arrival/departure information of the vehicle; and a server, which includes: a receiving module, which receives the The driving route of the vehicle and the GPS information and arrival/departure information of the vehicle transmitted by the vehicle-mounted mobile device. The driving route includes the first position and the second position; the statistical module is based on the GPS information and arrival/departure information. Station information counts the departure time at the first location of vehicles traveling on the driving route for consecutive shifts; the reading module receives the national roads of the national roads traveled by vehicles at each departure time from the first location to the second location Travel time; a historical database, which stores the historical travel time between the first position and the second position of the driving route for each vehicle at the departure time during the historical departure time/departure time; and the prediction module, which is Machine learning is used to predict the travel time of the vehicle at the next departure time from the first position to the second position of the driving route according to the departure time and the travel time of the national road and the historical travel time of each vehicle at the departure time. For example, the system described in item 1 of the scope of patent application, wherein the statistics module counts the departure time of the two consecutive shifts of the driving route at the first location based on the GPS information and arrival/departure information. Such as the system described in item 1 of the scope of patent application, wherein the reading module reads the departure time from the national highway electronic toll collection system based on the GPS information. The vehicle travels from the first location to the second location on the national highway National road travel time. For the system described in item 1 of the scope of patent application, the machine learning system is LSTM. For example, the system described in item 1 of the scope of patent application, wherein the LSTM structure includes: a normalization layer, which normalizes the national road travel time and the historical travel time of each vehicle at the departure time and each departure time ; The conversion layer is to convert the normalized data into the time-level data for training and learning of the LSTM; the LSTM model is to train and learn the time-level data by inputting two time levels and outputting one time level; The label retention rate of the training and learning is set to reduce the complexity of the LSTM model after the training and learning; and the fully connected layer is to output the reduced complexity of the LSTM model, so that the first order of the output is to predict the next departure Time is the travel time of the vehicle from the first position to the second position of the driving route. A method for predicting the arrival time of a vehicle includes: enabling a receiving module to receive the driving route of the vehicle and the GPS information and arrival/departure information transmitted by the vehicle-mounted mobile device on the vehicle, wherein the driving route includes a first position And the second position; Command the statistics module to count the departure time at the first location of vehicles traveling on the driving route in consecutive shifts based on the GPS information and arrival/departure information; order the reading module to read the departure time of each vehicle at the departure time from the first location The travel time of the national road from the location to the second location on the national road; order the historical database to provide the vehicle with the departure time in the history of the departure time/departure time, between the first location and the second location of the driving route Historical travel time; and make the prediction module use machine learning based on the national road travel time and the historical travel time of each vehicle at the departure time and each departure time, and then predict the next departure time from the first vehicle of the travel route Travel time from one location to the second location. For example, the method described in item 6 of the scope of patent application, wherein the statistics module is made to count the departure time of the two consecutive shifts of the driving route at the first location based on the GPS information and arrival/departure information. Such as the method described in item 6 of the scope of patent application, wherein the reading module is made to read the departure time from the national highway electronic toll collection system based on the GPS information. The vehicle travels from the first location to the second location on the national highway National road travel time. The method described in item 6 of the scope of patent application, wherein the machine learning system is LSTM. For example, the method described in item 9 of the scope of patent application, wherein the LSTM predicts the travel time of the vehicle at the next departure time from the first position to the second position of the driving route includes: using a normalization layer to divide the departure time and The national road travel time of each vehicle at the departure time and the historical travel time are normalized; Use the conversion layer to convert the normalized data into the time-level data for training and learning of the LSTM; use the LSTM model to input two time-levels and output one-time level to train and learn the time-level data; use the discard layer to define the training The learned label retention rate is used to reduce the complexity of the trained LSTM model; and the fully connected layer is used to output the reduced complexity of the LSTM model so that the first order of the output is to predict the next departure time of the vehicle from the The travel time from the first position to the second position of the driving route.

Images