TWI522974B - Arrival time prediction system and method - Google Patents

Description

Arrival time prediction system and method

The present invention relates to an arrival time prediction system and method, and more particularly to a system and method for predicting arrival time based on a stochastic neural network group.

At present, the conventional techniques for predicting the arrival time of public transport mainly use historical data to perform statistics and averaging, obtain the average speed and travel time between stations, or apply the instantaneous instantaneous speed information of the current vehicle. To estimate the arrival time. However, these methods are unable to reflect changes in the instantaneous road conditions between stations, resulting in large arrival time information errors.

Taiwan Patent Publication No. TW201137803 mainly proposes to collect the arrival information reported by the past bus to estimate the average speed and travel time between stations, and can be counted according to different weeks and time periods. When the user queries, the historical average can be given. Speed and travel time. Although this method can quickly provide the estimated arrival time, it is mainly based on the historical data average, and it is impossible to predict the arrival time due to the immediate road condition, so it may cause a large error in the arrival time prediction.

Taiwan Patent Publication No. TW201344647 first predicts the time after the station, according to the public The car's immediate location information, providing speed adjustment advice to the driver, in order to improve the on-site punctuality rate. Although this method can estimate the arrival time and provide on-site control, it is mainly based on the historical data average, and it is impossible to predict the arrival time due to the immediate road conditions, so it may cause a larger arrival time prediction. Error.

Taiwan Patent Publication No. TW201405497 mainly proposes that when the vehicle-mounted mobile equipment travels through various sections, the vehicle-mounted mobile equipment immediately returns the travel time of each section to the monitoring center of the back-end, and then the monitoring center records and publishes the shortest travel time and the longest travel of each section. Time is given to all car mobile devices. If the travel time of the car mobile device is between the shortest travel time and the longest travel time, it will not be returned. Although this method can effectively grasp the travel time of each road section and reduce the number of transmissions, it does not propose the prediction method of the bus arrival time, so it is impossible to predict the bus arrival information.

Taiwan Patent Publication No. TW201117146 mainly provides bus travel time inquiry, which allows users to check the instant location and travel time of the bus they want to board. Although this method allows the user to check the immediate location and travel time of the bus, it does not propose a method for predicting the arrival time of the bus, so it is impossible to predict the bus arrival information.

Taiwan Patent Publication No. TW200828190 mainly proposes to use the user's mobile device and receive the station information by the mobile device. When the site arrives, a notification is sent to remind the user. Although this method can alert the user when they arrive at the site, they can provide instant arrival information, but they cannot provide prediction information.

Taiwan Patent Announcement No. TWI252441 mainly proposes to receive the satellite positioning signal from the bus, and immediately return the location information to the monitoring center, and then the monitoring center provides the prediction module to predict the arrival time of the bus according to the instantaneous position of the bus. Although this method can provide a station time prediction, The patent mainly refers to the reference experience value, and the future method of forecasting the bus arrival time.

Taiwan Patent Publication No. TWI341998 mainly proposes to use the instantaneous speed of the bus and the distance to each station to predict the travel time; and to analyze the user's walking speed and the distance to each station, and calculate the walking time accordingly. Finally, estimate the suitable site based on travel time and walking time. Although this method can provide bus travel time prediction, it mainly considers the current speed and arrival distance of the bus. However, the traffic information between the vehicle and the station is not considered, so it may cause a larger arrival time prediction. Error.

Taiwan Patent Publication No. TW201232489 proposes to use the empirical mode decomposition method of Hilbert-Huang transform (HHT) to combine the gray mode to predict the driving speed, and then convert the estimated speed into travel time and arrival time. Although this method can effectively use the mathematical and statistical models for vehicle speed prediction, because it uses all the data for analysis, it cannot avoid the influence of extreme values, and it may cause a large error in the station time prediction.

In view of the above-mentioned problems of the prior art, the object of the present invention is to provide an arrival time prediction system and method for collecting travel time between stations and stations in various sections and time periods, and proposing a novel stochastic neural network. The group analyzes the aforementioned travel time data set, establishes a plurality of neural network models to avoid the influence of extreme values, and comprehensively considers the prediction results of a plurality of neural network models to improve the prediction accuracy, thereby predicting the user. The arrival time of the bus to be used is provided to the user for reference.

The arrival time prediction system of the invention comprises a plurality of station stop signs, a plurality of vehicle terminal devices, a plurality of cell network base stations, a cloud computing server, and a cloud history. A database and a plurality of inbound time prediction system client devices. Among them, each station stop sign has a latitude and longitude coordinate information. When the vehicle-mounted terminal devices approach the station stop signs, each of the vehicle-mounted terminal devices senses the latitude and longitude coordinate information, thereby generating the arrival station information. The arrival information is transmitted via the cell network base stations, and the cloud computing server receives the arrival information transmitted by the cell network base station, calculates the travel time, and predicts the remaining time according to the travel time and the query site. Travel time, and converted to the arrival time, and then the arrival time is transmitted via the cell network base station. The cloud history database stores latitude and longitude coordinate information and travel time between station stop signs. The arrival time prediction system client device sends the query site and receives the arrival time transmitted via the cell network base station, and then displays the arrival time.

The arrival time prediction method of the present invention comprises the steps of: setting a parameter value of a stochastic neural network group algorithm; reading a travel time between stations in a historical database; randomly generating m neural network models; Filter out the neural network model with the correct rate lower than the threshold, and leave the remaining k neural network models; obtain the instantaneous station-to-station travel time or test data in the test phase; travel time or test data Input to the filtered k-type neural network model, and predict the travel time of the station to the station; and obtain the predicted station-to-station travel time, converted to the arrival time of the target station.

In view of the above, the arrival time prediction system and method of the present invention may have one or more of the following advantages:

1. The present invention collects travel time between station-to-station for each road segment and time period in real time to estimate the travel time of the current vehicle location to the target site.

2. The present invention proposes a novel stochastic neural network group to analyze the aforementioned travel time data set, establish a plurality of neural network models, and comprehensively consider the prediction knots of a plurality of neural network models. In order to improve the accuracy of the prediction, it is predicted to predict the arrival time of the bus that the user wants to take, and provide the user with reference.

3. In the learning phase of the stochastic neural network group algorithm, the stochastic data is randomly selected from the data set for each type of neural network model as training data, and the remaining data is used as the training stage. In the test data, input training data to various types of neural network models for learning, to avoid the impact of extreme values.

4. In the test phase and the implementation phase of the stochastic neural network group algorithm, the present invention uses the weights averaged by the travel time predicted by each type of neural network model and the training phase, and finally the weighted average. The post travel time is used as the travel time prediction value of the stochastic neural network group algorithm, and the travel time is converted into the arrival time to perform the arrival time prediction.

100‧‧‧ Station stop sign

101‧‧‧Vehicle terminal equipment

102‧‧‧ cell network base station

103‧‧‧Cloud computing server

104‧‧‧Cloud computing room

105‧‧‧Cloud History Database

106‧‧‧Site time prediction system client device

S201~207, S301~S306‧‧‧ steps

1~8‧‧‧ class neural network model

Figure 1 is a schematic illustration of the arrival time prediction system of the present invention.

Figure 2 is a flow chart of the method for predicting the arrival time of the present invention.

Figure 3 is another flow chart of the method for predicting the arrival time of the present invention.

Fig. 4 is a schematic diagram of a neural network model such as the present invention.

Figure 5 is a schematic illustration of the predicted travel time of the present invention.

Referring to FIG. 1, the present invention relates to a system for arriving time prediction based on a stochastic neural network group. This system can mainly predict the arrival time of vehicles and is suitable for passenger transportation. , logistics operators, or other relevant operators who have the time to forecast the arrival time, and provide the predicted arrival time to the client device, so that the customer or user can instantly grasp the vehicle information and the arrival information, saving waiting time. It mainly consists of the following six modules: (1) Multiple station stop signs 100: This stop sign device mainly contains a set of latitude and longitude coordinate information, and this information can be pre-stored in the vehicle terminal device and the cloud computing server, when the vehicle terminal When the device approaches the station stop sign, the in-vehicle terminal device can sense the station information. In addition, the station device can also be embedded in a Radio Frequency IDentification (RFID) tag, which can sense the station card when the vehicle is approaching, and can determine the station accordingly. (2) A plurality of vehicle terminal devices 101: The device mainly includes a Global Positioning System (GPS) module, a cell network module, and a database module (not shown), which can collect the current position of the vehicle. (including latitude and longitude coordinates), and determine whether the current position is near the station stop sign 100, if the station is near the station stop sign 100, the station is judged, and the arrival station information and time point are transmitted back to the cloud operation via the cell network base station 102. Server end 103. In addition, in the part of the station judgment, the in-vehicle terminal device 101 can also be embedded in the RFID reader, and can sense the station card when the vehicle is approaching, and can receive the RFID tag signal from the station card device to determine whether to arrive at the station. (3) a plurality of cell network base stations 102: each cell network base station 102 can provide data transmission and reception functions, and is responsible for transmitting the vehicle terminal device 101, the cloud computing server 103, and the arrival time prediction system client. Data transfer between devices 106. (4) Cloud computing server 103: This server can mainly collect and analyze the arrival information from the vehicle terminal device 101, the arrival time point, and calculate the travel between each station and the station according to each arrival time point. At the time, the data set of the travel time between the plurality of stations and the station before the travel route of the target site of the user query is input to the training method of the stochastic neural network group arrival time prediction method proposed by the present invention. Such a neural network group performs analysis and calculation to obtain the target site The remaining travel time forecast is then converted to the arrival time at the target site. (5) Cloud History Database 105: This database can store the travel time between each station and station in history. It can be used as a training data set of random neural network groups to train each type of nerve. Network model. (6) a plurality of arrival time prediction system client device 106: the device can be a mobile device with a human-machine interaction interface and a network transmission module, which allows the user to query and display the desired device through the device. The arrival time prediction of the target site. The user can pre-set the site and time to be boarded by the user, and then the device actively updates and judges, and promptly sends a reminder message and sound to the user when the vehicle is about to arrive.

Referring to FIG. 2 and FIG. 3, the present invention further provides a method for arriving time prediction based on a stochastic neural network group. This method will mainly consist of two phases: (a) training phase and (b) implementation and testing phase. The training phase mainly includes four steps: step S201: setting a random neural network group algorithm parameter value; step S202: reading a travel time between each station in the historical database; S203: randomly generate m neural network models; and step S208: filter out a neural network model whose correct rate is lower than a threshold, and leave k neural network models. The implementation and testing phase mainly includes three steps, namely: step S301: obtaining real-time travel time in each station-to-station or test data in the test phase; step S302: inputting data to the filtered k classes The neural network model predicts the travel time between the stations and the station; and step S306: after obtaining the predicted station-to-station travel time, the time of arrival is converted to the target station.

In step S201, first, the developer of the arrival time prediction system sets the relevant parameter values of the stochastic neural network group algorithm, including the number of neural network models (the following description will take m as an example), the neural network The maximum number of hidden layers in the road model (the following description will take h _max as an example), the maximum number of neurons in each hidden layer in the neural network model (the following description will take c _max as an example), the training neural network The ratio of the training data of the model to the total number of training data (the following description will take r% as an example) and the correct rate threshold (the following description will take w _threshold as an example).

In step S202, the time when the vehicle arrives at each station is obtained from the cloud history database 103, and converted into the travel time between the stations, for example, the arrival time of the station 1 is the time point t1, and the station 2 is When the arrival time is time point t2, the travel time of station 1 to station 2 is |t2-t1|. Then, the travel time set is used as the input and output data of the neural network model for subsequent learning. Taking Figure 1 as an example, if you want to predict the time of arrival at station n when the vehicle is traveling to station n-2 (ie, the travel time of the target output is |tn-tn-2|), the input travel time data set may contain {|t2 -t1|,|t3-t2|,...,|t n-2-tn-3|}.

In step S202, m neural network models are randomly generated according to the parameter values of the stochastic neural network group algorithm set by the station time prediction system developer, and each of the neural network models is randomly obtained. The r% of the number of training materials is used for training and learning, and the remaining data (ie, 100%-r% of the data amount) is used as the verification of each type of neural network model, and each type of neural network model will be obtained. Different materials are trained and verified. In addition, each neural network model will generate 0~hmax hidden layers according to the parameter settings, and generate 0~cmax neurons for each hidden layer, and the hidden layer of each type of neural network model and The combination of neurons will be different. Then input the above r% data into the neural network Training and learning in the model, after reaching convergence, input 100%-r% of the data (ie the test data in the training phase) into the trained neural network model, and obtain the predicted travel time, and the correct The travel time is compared, and the correct rate of each type of neural network model is obtained, and the correct rate is used as the weight value in the implementation and test phases.

In step S208, the neural network model with the correct rate lower than the threshold value is filtered out, and the remaining k neural network models are used: the correct rate and the correct rate threshold of the m neural network models generated randomly. _{The threshold} is compared, and the neural network model below the threshold (that is, the correct rate is too low) is excluded, and k neural network models are left; if there is no neural network model, the correct rate is high. When the threshold is exceeded, the process returns to step S201, where the developer of the station time prediction system resets the threshold and retrains the random neural network group.

In step S301, in the implementation and testing phase, the travel time between the station and the station of the instant vehicle is first obtained, for example, the vehicle is moved to the station n-2 in FIG. 1, and the user wants to query the station n. The arrival time prediction (ie, the travel time of the target output is |tn-tn-2|). At this point, the travel time data set {|t2-t1|,|t3-t2|,...,|t n-2-tn-3|} of the vehicle in this journey can be calculated and this will be Input data as a neural network model.

In step S302, the instant travel time data set {|t2-t1|, |t3-t2|,..., |t n-2-tn-3|} is obtained and input to the filtered k-like nerves. In the network model, each neural network model predicts a predicted travel time of |tn-tn-2|, which is then multiplied by the weight values of the various neural network models obtained by the training phase ( That is, the correct rate of each type of neural network model in the training phase), and the weighted value is summed by the sum of the weight values (ie, weighted average).

In step S301, after obtaining the predicted travel time |tn-tn-2| obtained by comprehensively considering the k neural network models, the predicted travel time is added according to the instantaneous time point tn-2 of the vehicle|tn- Tn-2| obtains the arrival time prediction of the station n and provides the prediction result to the user.

The invention collects and analyzes the information of the arrival (departure) from the in-vehicle terminal device 101 (including site information and time points, etc.), and converts the data collection into station-to-station travel time and stores it in the cloud history data. In the library 105, and in the cloud computing server 103, an on-site information prediction method module based on a stochastic neural network group algorithm is designed and implemented, and the travel time set in the cloud history database 105 can be accessed, and According to this, input to the station-based information prediction method module based on the stochastic neural network group algorithm, and perform neural network model training to predict travel time. When the station information prediction system client performs the site-to-station time prediction, the current vehicle terminal device 101 can input the plurality of site information into the trained neural network group before the return of the route to reach the target station. The travel time prediction of the point, re-converting and providing the arrival time to the target site to the station information prediction system client device 106. The technical feature of the present invention is mainly to propose and design a stochastic neural network group algorithm and apply it to the station information prediction. The following description will be made with an embodiment.

The present invention provides a system for arriving time prediction based on a stochastic neural network group, the system architecture of which is shown in FIG. The system will include a plurality of station stop signs 100, a plurality of in-vehicle terminal devices 101, a plurality of cell network base stations 102, a cloud computing server 103, a cloud history database 105, and a plurality of inbound time prediction system clients. End device 106. In the present embodiment, the station stop sign 100 of the same route is taken as an example. There are n stations in the route, and each station has location information (including longitude and latitude). As shown in Table 1, there are 12 stations in total in route 1 (that is, n in Figure 1 is 12), and the corresponding latitude and longitude can be stored in the vehicle-mounted terminal equipment; when the vehicle number 1 is from the station 1 to the station 2, At 2014/4/1 14:53, the GPS module of the vehicle terminal device detected that the vehicle was 120.97839 and the latitude was 24.808658. When the station 2 is approached (for example, within a straight line distance of 30 meters), it is determined that the station is arriving, and the arrival information (including the station number and time point) is transmitted back to the cloud computing server 103 via the cell network base station 102.

In addition, the station stop sign 100 may also be provided with an RFID tag, and the in-vehicle terminal device 101 may be provided with an RFID reader, and when the in-vehicle terminal device 101 is adjacent to the station stop sign 100, the RFID tag of the station stop sign 100 may be detected, and This determination is to arrive at the station, and then the arrival information (including the station number and time point) is transmitted back to the cloud computing server 103 via the cell network base station 102. As shown in Table 2, the vehicle arrival information return data collection will mainly record the route number, vehicle number, station number, and time point, and the cloud computing server 103 can convert the vehicle arrival information into station arrival time. Information (as shown in Table 3), and the information is stored in the cloud history database 105. For example, the time when the vehicle number 1 departs from the station 1 is 2014/4/1 14:46:28, and arrives at the station 2 at 2014/4/1 14:53:31, so the travel time from station 1 to station 2 is 423 seconds; the time when the vehicle number 2 departs from the station 1 is 2014/4/1 19:32:22, and arrives at the station 2 at 2014/4/1 19:40:13, so the trip from station 1 to station 2 The time is 471 seconds.

When there is a vehicle number 10001 driving to the station 6 (ie, the cloud server 103 knows the travel time between the station and the station between the station 1 and the station 6), and there is a station time prediction system client device to the cloud computing server 103 Query route number 1 The arrival time of station 12 (i.e., predict the travel time of station 6 to station 12 and convert to the arrival time of station 12). At this time, the cloud computing server 103 can use the data in the cloud history database 105 (ie, the station-to-station travel time information of the route numbers 1 and 2, as shown in Table 4) as a stochastic neural network group algorithm for training. Stage data to establish a stochastic neural network group and use this algorithm to predict the arrival time.

Table 1 Station Location Information

The method for predicting the arrival time based on the stochastic neural network group of the present invention has a method flow as shown in FIGS. 2 and 3. This method will mainly consist of two phases: (a) training phase and (b) implementation and testing phase.

The training phase mainly includes four steps, which are respectively step S201: setting a random neural network group algorithm parameter value; S202: reading a travel time between each station in the historical database; S203: randomly generating m a neural network model; and S208: filtering out a neural network model with a correct rate lower than the threshold, and leaving k neural network models.

The implementation and testing phase mainly consists of three steps, namely S301: obtaining real-time test data between each station-to-station travel time or test phase; S302: inputting data to the filtered k-like neural networks The model, and predicts the travel time between the stations and the station; and S306: after obtaining the predicted station-to-station travel time, converted to the arrival time of the target station.

In the training phase, the relevant parameter values of the stochastic neural network group algorithm are first set by the arrival time prediction system developer (step S201). For example, set a total of 10 neural network models (ie, m is 10), the maximum number of hidden layers in the neural network model is 5 (ie, h _max is 5), and the maximum neural layer of each hidden layer in the neural network model. The number of elements is 7 (ie, c _max is 7), the training data of the training neural network model accounts for 60% of the total training stage data (ie, r% is 60%), and the correct rate threshold is 0.945 ( That is, the w _threshold is 0.945=94.5%), and 10 neural network models will be generated according to this parameter value to perform the arrival time prediction.

In this step S202, the time when the vehicle that has obtained the history from the cloud history database arrives at each station is converted into the travel time between the stations and the stations, as shown in Table 4. Since in the present embodiment, the vehicle to be predicted travels to the station 6, and the arrival time of the station 12 is to be predicted, and the arrival time data set between the stations 1 to 6 is known {t1, t2, t3, t4, T5, t6}, converted into a travel time data set between station and station {|t2-t1|, |t3-t2|, |t4-t3|, |t5-t4|,|t6-t5|}, and To predict the travel time of station 6 to station 12 (ie, the travel time of the target output is |t12-t6|). In the present embodiment, the travel time data set {|t2-t1|, |t3-t2|, |t4-t3|, |t5-t4|, |t6-t5|} are named as parameter names {x1, x2, respectively. , x3, x4, x5}, and the travel time of the target output |t12-t6| is named as the parameter name y.

Step S203 randomly generates m neural network models, and further includes step S204: generating training data and verification data. In detail, the present invention randomly generates 10 neural network models based on the parameters of the stochastic neural network group algorithm set by the station time prediction system developer, and sets the maximum number of hidden layers in the neural network model. 5. The number of neurons in each hidden layer in the neural network model is 7, which means that the number of hidden layers in each neural network model will be between 0 and 5, and the number of neurons in each hidden layer will be In the case of 0 to 7, the implementation of the result is shown in Table 5 (step S205). The hidden layer of the neural network model 1 is 1 layer, and the number of neurons in the hidden layer is 2 (as shown in Fig. 4); the hidden layer of the neural network model 2 is 2 layers, and the first layer is hidden. The number of neurons in the layer is 3, and the number of neurons in the hidden layer in the second layer is 4; and so on, 10 neural network models can be obtained. Moreover, since the training data of the training-like neural network model accounts for 60% of the total number of data in the training phase, taking Table 4 as an example, the total number of data in the training phase is 10,000, so each type of neural network model Randomly take 6,000 pens as training neural network model learning The test data in the remaining 4000 training stages (Testing Data in TRaining Stage, TTDRS) will be used as the verification of each type of neural network model in the training phase. In this step, each set of 6,000 data obtained by each type of neural network model is randomly generated, and each type of neural network model will acquire different sets of data for training and learning.

Step S206: training and learning of a neural network model. In this embodiment, 10 neural network models will input 6,000 data for training and learning respectively. The following uses a neural network model 1 (as shown in Fig. 4) as an example, in which a neural network is used. The 6000 data of the road model 1 is a data combination including the route number 1 and the route number 10000, and the training and learning description of the neural network model 1 is as follows.

Step i: randomly generate the weights of the individual neurons, and the constant terms of the hidden layer and the output layer neurons, as shown in Table 6.

Step ii: Input 6,000 pieces of data into the neural network model 1 one by one, and take the route number 1 as an example. First, the data is normalized to a value between 0 and 1, so the data in the embodiment are all less than 5000, so the normalization is divided by 5000, and the results are shown in Table 7. Then, according to the input signal, the output signal of each hidden layer neuron is calculated, wherein the embodiment uses Logistic allocation (ie, The way to calculate the output signal is as follows.

Neuron 6: Total input signal:

Convert output signal:

Neuron 7: Total input signal:

Convert output signal:

Step iii: Calculate the output signal of the output layer neurons according to the hidden layer output signal.

Neuron 8: Total input signal:

Convert output signal:

Step iv: Compare the error terms of the output value (ie 0.759554) with the true value (ie 0.7796).

Neuron 8 error term:

Step v: The error term is fed back to the hidden layer, and the error term of the hidden layer neuron is calculated separately.

Neuron 6 error term:

Neuron 7 error term:

Step vi: Update each neuron weight and constant term according to the neuron error term, and set the learning rate σ to 0.8 in this embodiment.

Repeat steps ii ~ learning step VI, each piece of data is input to the neural network model, this turn until the output signal of the difference output signal of a round lower than a threshold value o _threshold (in: step vii In this example, the o _{threshold is} set to 0.01), then convergence is achieved and learning is completed, and the individual weights and constant terms of the neural network model are determined.

The foregoing is a training and learning process of the neural network model 1, and at the same time, training other neural network models (ie, the neural network model 2~ neural network model 10) can support parallel operations. After the completion of the training, the follow-up steps ii to iii may be repeated in the prediction of the travel time between the station 6 and the station 12, and the test data or the real-time data is used as the input signal, and the output signal is the predicted value during the travel. Among them, the travel time predicted value produced by the neural network model needs to be restored again, and the travel time is obtained. For example, the output signal is 0.759554, and the travel time is 3797.769233 seconds.

Step S207: class-like neural network model verification and weighting. When all the gods are finished After training and learning through the network model, the remaining 4000 data can be used to verify each type of neural network model, and the average correct rate is calculated as the weight of each type of neural network model. Taking the neural network model 1 as an example, all the test data in the training phase are input into the trained neural network model 1 and steps ii to iii are repeated to calculate the correct rate. For example, when the route number 10000 is an input signal, the normalized value is as shown in Table 8. The predicted value is 0.75986369, and the predicted value is multiplied by 5000 to 3799.318449. The correct rate is 1-(|true value- Predicted value|/true value=1-(|3939-3799.318449|/3939)=96.45%; and so on, the average correct rate of test data (TDTRS) in 4000 training stages can be calculated, in this case 93.23 %. In this embodiment, the average correct rates corresponding to the 10 neural network models are 93.23%, 94.90%, 94.03%, 93.57%, 94.61%, 93.52%, 94.93%, 95.21%, 94.48%, and 94.45%, respectively. As shown in Table IX.

Step S208: Filter out the neural network model with the correct rate lower than the threshold, and leave the remaining k neural network models. This step will analyze the average correct rate of each type of neural network model, and filter out the _{threshold of the} correct rate _threshold (ie, 94.5% set in this embodiment), where the neural network model 1, the neuron Network model 3, neural network model 4, neural network model 6, neural network model 9, neural network model 10, etc. will be filtered out, leaving 4

The neural network model and its weight values are used for the implementation and testing phases.

In step S301, when the implementation and testing phases are taken, the instant vehicle arrives at the station. The information is input to the trained stochastic neural network group to perform the arrival time prediction. For example, the arrival time prediction system client device will check the arrival time of the arrival station 12 at 2014/5/3 11:59:00, and will take the arrival time of the station 1~ station 6 and the travel between the station and the station. The time (as shown in Table 11), as the input data of the stochastic neural network group (as shown in Table 12), obtains the travel time of the target predicted value station 6 to station 12.

In addition, the arrival time prediction system developer can also collect historical data as test data (TDTES) in the test phase, and obtain the travel time between each station and station of each route number as a random neural network group. Enter values to analyze and optimize stochastic neural network populations.

In step S302, the data is input to the filtered k-type neural network models, and the travel time between the stations is predicted. After obtaining the input data, the data can be used as each filtered neural network model (ie, neural network model 2, neural network model 5, neural network model 7, neural network model 8, As shown in Table 10, the input signals are predicted by the neural network model 2, the neural network model 5, the neural network model 7, and the neural network model 8 to predict the travel time of 3766.607 seconds and 3587.98 seconds. 3661.828 seconds, 3724.095 seconds (step S303), as shown in Table 13. Finally, the weighted average is calculated according to the weight of each type of neural network model (steps S304~S305) to obtain the travel time predicted value of 3752.516552 seconds (ie [94.90% * 3766.607 + 94.61% * 3857.98 + 94.93% * 3661.828 + 95.21% * 3724.095]/[94.90%+94.61%+94.93%+95.21%]=3752.516552).

In step S306, after the predicted station-to-station travel time is obtained, it is converted into the arrival time of the target station. After obtaining the station-to-station travel time prediction value, the current arrival information can be converted into the arrival time of the arrival target station based on the station-to-station travel time prediction value. The time point at which the route number 10001 of the present embodiment arrives at the station 6 is 2014/5/3 11:58:46, and the predicted travel time of the station 6 to the station 12 is 3752.516552 seconds, so the station 12 predicts that the station time is 2014/ 5/3 13:01:19, then pass this information back to the arrival time prediction system client device.

The actual application to the passenger transport industry is based on the data of passenger transporter A. The total data of the entire month of March 2014 is collected, including 2,956 baht. The experimental environment covers 40 road sections and uses different Data exploration algorithm to test its positive The accuracy rate includes Logistic Regression (LR), the traditional Back-Propagation Neural Network (BPNN), and the random neural network group proposed by the present invention (Random Neural Networks, RNN), confirming that this method is indeed superior, the experimental results are shown in Table 14.

In summary, the stochastic neural network group-based arrival time prediction system and method of the present invention collects travel time between stations and stations in each road segment and time period, and proposes a novel stochastic neural network group. To analyze the aforementioned travel time data set, establish a plurality of neural network models to avoid the influence of extreme values, and comprehensively consider the prediction results of a plurality of neural network models to improve the prediction accuracy, thereby predicting the user's desire The arrival time of the bus will be provided to the user for reference.

The above is intended to be illustrative only and not limiting. Any equivalent modifications or alterations to the spirit and scope of the invention are intended to be included in the scope of the appended claims.

100‧‧‧ Station stop sign

101‧‧‧Vehicle terminal equipment

102‧‧‧ cell network base station

103‧‧‧Cloud computing server

104‧‧‧Cloud computing room

105‧‧‧Cloud History Database

106‧‧‧Site time prediction system client device

Claims (5)

An arrival time prediction system includes: a plurality of station stop signs, each station stop sign has a latitude and longitude coordinate information; and a plurality of in-vehicle terminal devices, when approaching the station stop signs, the in-vehicle terminal devices sense the The latitude and longitude coordinates information, thereby generating a plurality of arrival information; a plurality of cellular network base stations, the arrival information is transmitted through the cellular network base stations; and a cloud computing server receives the cellular networks The base station transmits the received arrival information, calculates a plurality of travel times, and then predicts a remaining travel time based on the travel time and a query site and converts to a stop time, and the arrival time is Transmitting through the cell network base station; a cloud history database storing the latitude and longitude coordinate information and the travel time between the station stop signs; a plurality of arrival time prediction system client devices, transmitting the Querying the station, and receiving the arrival time transmitted through the cell network base stations, and then displaying the arrival time; wherein the cloud computing servo Department use a neural network to predict the population of some remaining travel time; and wherein the cloud database to store the history of these travel Time, as a training data set for training the stochastic neural network group, and training a plurality of neural network models. According to the arrival time prediction system of claim 1, wherein each of the vehicle terminal devices further comprises: a global positioning system (GPS) module, which collects location information of each of the vehicle terminal devices, and according to the latitude and longitude Coordinate information and the location information determine whether to arrive at the station, thereby generating a plurality of arrival time and the arrival information; a cell network module, transmitting the arrival time and the information of the stations to the cell networks At least one of the base stations; a database module for storing the latitude and longitude coordinate information, wherein each of the station stop cards is embedded with an RFID tag, and each of the vehicle terminal devices is embedded in an RFID reader When each of the in-vehicle terminal devices is adjacent to each of the station stop signs, each of the RFID readers senses each of the RFID tags, and accordingly determines whether or not to arrive at the station. An arrival time prediction method includes: setting a plurality of stochastic neural network group algorithm parameter values; reading a plurality of travel times between stations in a historical database; randomly generating m neural networks Model; filtering out the neural network model with the correct rate lower than the threshold, and leaving the remaining k neural network models; Obtaining a plurality of test data during the travel time or test phase between the station and the station; inputting the travel time or test data into the filtered k-type neural network model, and predicting the station-to-station a plurality of travel times; and an arrival time of the target station after the predicted travel time of the station is obtained; and wherein the stochastic neural network group algorithm parameter values include a neural network model The number, the maximum number of hidden layers in a type of neural network model, the maximum number of neurons in each hidden layer in a type of neural network model, the number of training data in a training-type neural network model, the proportion of data in the total training phase, and one The correct rate threshold. According to the method for predicting the arrival time of item 3 of the patent application scope, wherein the filtering of the neural network model with the correct rate lower than the threshold value, and the remaining k neural network models include: randomly generating m The correct rate of the neural network model is compared with the correct rate threshold, and the neural network model below the correct rate threshold is excluded. According to the method for predicting the arrival time of item 3 of the patent application scope, wherein the travel time or test data is input to the filtered k-type neural network model, and the step of predicting the plurality of travel times of the station-to-station is performed. The method further comprises: obtaining an instant travel time data set; inputting the travel time data set to the filtered k-type neural network model; A plurality of predicted travel times are predicted, which are then multiplied by a weight value of each of the neural network models; and the weighted values are summed by the sum of the weight values.