A method for predicting daily traffic flow parameters integrating the impact of holidays

CN113971479BActive Publication Date: 2024-11-26浙江省机电设计研究院有限公司
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202111167003.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-01
Publication Date
2024-11-26
Estimated Expiration
2041-10-01

AI Technical Summary

Technical Problem

The existing traffic flow prediction methods fail to effectively consider the impact of holidays on daily traffic flow parameters, and lack consideration of the differences in traffic flow conditions between holidays and non-holidays, resulting in insufficient prediction accuracy.

Method used

The method of combining convolutional neural network (CNN) and long-term memory network (LSTM) is used to obtain traffic flow parameter data, environmental factors and holiday label data, extract day and intraday traffic flow patterns, and pass the daily velocity data difference rate and Special labels for holidays are used for fusion prediction.

Benefits of technology

It significantly improves the accuracy of traffic flow parameters prediction on holidays, is better than many existing benchmark methods, and can effectively capture the correlation between traffic flow parameters and weather and holidays, providing more accurate traffic management reference.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113971479B_ABST
    Figure CN113971479B_ABST
Patent Text Reader

Abstract

The present invention provides a method for predicting daily traffic flow parameters that integrates the impact of holidays. Based on a deep learning method composed of a convolutional neural network (CNN) and a long short-term memory network (LSTM), the traffic flow pattern is combined with various environmental factors, while considering the impact of holidays on the traffic flow pattern, and integrating the "special or not" characteristics of holidays extracted from the traffic flow parameter data itself, to achieve traffic flow parameter prediction with a time series length of a whole day. The present invention has excellent prediction effect. Example analysis shows that the method has high prediction accuracy and strong operability, and is superior to many existing methods.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a method for predicting daily traffic flow parameters integrating holiday impacts, and belongs to the field of intelligent transportation. Background Art

[0002] Traffic flow prediction is an important issue in intelligent transportation, and has a research history of more than 20 years. Predicting traffic flow parameters (flow, speed) can provide theoretical support for the planning and design of transportation facilities, and also provide a reference for traffic management and control. Among them, the role of daily traffic flow parameter prediction is mainly to help traffic management agencies arrange appropriate management or signal control plans in a timely manner, thereby ensuring operational efficiency.

[0003] The daily traffic flow conditions on holidays (including national holidays and the day before and after national holidays) are often quite different from those on non-holidays. Therefore, it is of great significance to capture the correlation between traffic flow parameters and holidays. However, in recent years, most studies on traffic flow prediction are based on simple deep learning methods and do not consider the relationship between traffic flow parameters and holidays. In addition, there is not necessarily a significant difference between the daily traffic flow conditions on holidays and non-holidays. The daily traffic flow conditions on holidays may be very similar to those on non-holidays. Current research also lacks consideration of this point. Summary of the invention

[0004] In order to solve the problems existing in the background technology, the present invention proposes a method for predicting daily traffic flow parameters that integrates the influence of holidays. The basic idea of ​​this method is to obtain the traffic flow parameter data and the labeled data of environmental factor labels, holiday labels, and whether the holiday is special or not, and use a specific convolutional neural network (CNN) to extract the daytime and intraday traffic flow patterns from the traffic flow parameter data, and then input the extracted features into the long short-term memory network (LSTM) unit, integrate the labeled data of environmental factor labels, holiday labels, and whether the holiday is special or not, and obtain the final prediction result.

[0005] The basic steps of this method are as follows:

[0006] A method for predicting daily traffic flow parameters integrating the impact of holidays includes the following steps:

[0007] c1. Collect traffic flow parameter data;

[0008] c2. Label environmental factors;

[0009] c3. Mark holiday labels; the holidays include national holidays, the day before national holidays, and the day after national holidays;

[0010] c4. Define an indicator that characterizes the degree of difference between traffic conditions on holidays and non-holidays, namely, the daily speed data difference rate S; S is the percentage of time intervals with an absolute percentage error exceeding a certain threshold in the total number of daily time intervals when the daily speed data of any holiday is compared with the average daily speed data of all non-holidays within a given time range, assuming that the traffic flow parameter is selected as speed;

[0011] c5. Calculate the daily speed data difference rate S and mark whether it is a special holiday or not;

[0012] c6. Construct a daily traffic flow parameter prediction model;

[0013] c7. Predict daily traffic flow parameters.

[0014] In the above technical solution, further, the process of step c1 includes:

[0015] The traffic flow parameters are flow rate and speed, and the data are N consecutive days of data for two years or more; if the speed is to be predicted, only the speed data can be collected, and if the flow rate is to be predicted, both the flow rate and speed data need to be collected;

[0016] The method for collecting traffic flow parameter data is to record the vehicle flow or average speed at the data collection location at a certain time interval, and the time interval can be 5-30 minutes.

[0017] Furthermore, the process of step c2 includes:

[0018] The environmental factors include: day of the week, month, weather;

[0019] The method for marking the environmental factor label is:

[0020] The labeling method of the "day of the week" label is to find the day of the week for each day according to the date of the traffic flow parameter data obtained. If it is Monday, Tuesday to Sunday, it is labeled with 1, 2 to 7 respectively;

[0021] The labeling method of the "month" label is to find out which month each day belongs to in the twelve months of the year according to the date of the traffic flow parameter data obtained. If it belongs to January, February to December, the label is marked with 1, 2 to 12 respectively;

[0022] The marking method of the "weather" label is to find the weather conditions of each day according to the date of the traffic flow parameter data that has been obtained, and the corresponding ones are sunny, cloudy, overcast, light rain, moderate rain, heavy rain, light snow, moderate snow, heavy snow, and others, corresponding to 1, 2 to 10 respectively. Since the weather is time-varying, it needs to be marked every 2 hours. The marking of each day is carried out in the time sequence of 0:00-2:00, 2:00-4:00, ..., 22:00-24:00, that is, for each day, the label is marked with a one-dimensional vector of length 12.

[0023] Furthermore, the process of step c3 includes:

[0024] The holiday labels include: "Is it a national holiday?" label, "Is it the day before a national holiday?" label, and "Is it the day after a national holiday?" label;

[0025] The method for marking holiday labels is:

[0026] Among them, the labeling method of "Whether it is a national holiday or not" is to find out whether each day is a national holiday according to the date of the traffic flow parameter data that has been obtained, and mark the label as 1 or 2 if it is a national holiday or not;

[0027] The labeling method of "Whether it is the day before a national holiday" is to find out whether each day is the day before a national holiday according to the date of the traffic flow parameter data obtained. If it is the day before a national holiday, and if it is not the day before a national holiday, the label is marked with 1 and 2 respectively;

[0028] The labeling method of "Whether it is the day after a national holiday" is to check whether each day is the day after a national holiday according to the date of the traffic flow parameter data that has been obtained. The day after a national holiday and the day after a non-national holiday are marked with 1 and 2 respectively.

[0029] Furthermore, in step c4: the daily speed data difference rate S is an indicator for judging the degree of difference between the traffic conditions of any holiday and non-holiday within a given time range, and its value is between 0 and 1. The closer it is to 0, the smaller the difference between the traffic conditions of this holiday and non-holiday is, and the closer it is to 1, the greater the difference between the traffic conditions of this holiday and non-holiday is.

[0030] Furthermore, in step c5, the daily speed data difference rate S is calculated as follows:

[0031] 1) Filter non-holidays within a given time range to obtain all non-holiday daily speed data, with a total of m time intervals per day; calculate the average speed data of all non-holidays in the same time interval to obtain the average daily speed data data_avg of non-holidays;

[0032] data_avg=[data_avg_1,data_avg_2,...,data_avg_m]

[0033] 2) For any holiday d within a given time range h , whose daily speed data is data_d h ;

[0034] data_d h =[data_1,data_2,...,data_m]

[0035] 3) Calculate the absolute percentage error between the speed data of the holiday and the average daily speed data of non-holidays at each time interval, and obtain the daily speed absolute percentage error vector difference_d h ;

[0036]

[0037] difference_d h =[difference_1,difference_2,...,difference_m]

[0038] Wherein, difference_i is the absolute percentage error of speed in the i-th time interval;

[0039] 4) Set the threshold threshold1;

[0040] 5) Statistic difference_d h The number of difference_i greater than threshold1 is denoted as c, and the daily speed data difference rate S is:

[0041]

[0042] Furthermore, in step c5, the method of marking whether a holiday is special or not is as follows:

[0043] For each holiday within a given time range, the corresponding daily speed data difference rate S is calculated, and a threshold threshold2 is set. If the corresponding S is greater than threshold2, the holiday is considered special and belongs to a special holiday. Otherwise, the holiday is considered not special and belongs to a non-special holiday.

[0044] The labeling method of "Is the holiday special or not" is to mark each holiday with 1 if it is a special holiday or 2 if it is a non-special holiday.

[0045] Furthermore, step c6 is specifically as follows:

[0046] c61. Build a daily traffic flow parameter prediction model that integrates convolutional neural network (CNN) and long short-term memory network (LSTM);

[0047] c62, input data, the first part of the input data is the traffic flow parameter data obtained in step c1, the second part of the input data is the environmental factor label data obtained in step c2, the holiday label data obtained in step c3, and the holiday special or not label data obtained in step c5;

[0048] c63. Rationally divide the acquired traffic flow parameter data and label data into training set and test set;

[0049] c64. Reconstruct the input data of the first part, and reconstruct the traffic flow parameter data into a square matrix with a side length of M*7. M*7 is the time window of the convolution layer. The convolution layer spans the daytime and intraday dimensions. In the daytime dimension, M*7 represents M*7 consecutive days of training data. In the intraday dimension, it represents M*7 time intervals, and M is the optimal week number determined by the input specific traffic flow parameter data; the schematic diagram of the reconstructed traffic flow parameter data is shown in the figure attached to the specification. Figure 5 As shown, t in the figure is the time interval for recording traffic flow parameter data.

[0050] Further explanation of the reconstruction method of input data: If M = 4, and the first day selected is January 1, then the columns of each input element are from the first column of January 1 to the M*7th column of January 28, and the time range included in each column is 28*t. The last row of traffic flow parameter data of the first input element is the traffic flow parameter data of the first time interval t of the corresponding date, the last two rows of traffic flow parameter data of the second input element are the traffic flow parameter data of the first and second time intervals t of the corresponding date, and the traffic flow parameter data of the last input element is the traffic flow parameter data of the last M*7 time intervals t of the corresponding date. The unfilled cells in the matrix are filled with the average traffic flow parameter data of the corresponding date.

[0051] As shown in the attached figure of the instruction manual Figure 6 The figure shows the input diagram of traffic flow parameter data, which includes m matrices, all of which are inputs of the convolutional layer.

[0052] c65, the convolutional neural network extracts vectors containing daytime and intraday traffic flow parameter information from the reconstructed traffic flow parameter data, and the long short-term memory network performs time series prediction on the output of the convolutional neural network;

[0053] c66. Perform one-hot encoding on the input data of the second part, that is, convert the labeled data of each label into a binary sequence;

[0054] c67. Combine the time series prediction results output by the long short-term memory network with the labeled data of the seven labels converted into one-hot encoding to output the final traffic flow parameter data time series prediction results;

[0055] c68. Use the test set data to verify the model.

[0056] Furthermore, step c7 is specifically as follows:

[0057] c71. For each holiday to be predicted, find the "holiday special or not" label of the corresponding historical day, and mark it with the label data with the highest frequency in the historical day labels. If the proportion of 1 and 2 in the historical day labels is the same, then the day is randomly marked as 1 or 2;

[0058] c72. Use the constructed model to predict daily traffic flow parameters; use the traffic flow parameter data and label marked data of the previous M*7 days, as well as the label marked data of the target prediction day, to predict the daily traffic flow parameters of that day.

[0059] The present invention has the following beneficial effects:

[0060] The present invention has significant practical significance. Each label is interpretable in reality. At the same time, the correlation between traffic flow parameters and environmental factors such as weather, the correlation between traffic flow parameters and holidays, and the differences within holidays are taken into account.

[0061] The present invention has excellent prediction effect. Example analysis shows that the method has high prediction accuracy, is better than many existing benchmark methods, and shows obvious prediction advantages on some holidays.

[0062] The present invention has strong operability, and the marking data of each tag is relatively easy to obtain. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is the overall framework of the method of the present invention;

[0064] Figure 2 Label the box diagram for environmental factors;

[0065] Figure 3 Label the boxes for the holiday labels

[0066] Figure 4 Label the boxes to indicate whether holidays are special or not;

[0067] Figure 5 This is a schematic diagram of the reconstructed traffic flow parameter data;

[0068] Figure 6 It is a schematic diagram of inputting traffic flow parameter data;

[0069] Figure 7 is the model block diagram;

[0070] Figure 8 is the prediction box diagram;

[0071] Fig. 9 This is a comparison chart of the effects. DETAILED DESCRIPTION

[0072] As shown in the attached figure of the instruction manual Figure 1 The figure shows the overall block diagram of the method of the present invention. A method for predicting daily traffic flow parameters integrating the impact of holidays comprises the following steps:

[0073] c1. Collect traffic flow parameter data;

[0074] c2. Label environmental factors;

[0075] c3. Mark holiday labels; the holidays include national holidays, the day before national holidays, and the day after national holidays;

[0076] c4. Define an indicator that characterizes the degree of difference between traffic conditions on holidays and non-holidays, namely, the daily speed data difference rate S; S is the percentage of time intervals with an absolute percentage error exceeding a certain threshold in the total number of daily time intervals when the daily speed data of any holiday is compared with the average daily speed data of all non-holidays within a given time range, assuming that the traffic flow parameter is selected as speed;

[0077] c5. Calculate the daily speed data difference rate S and mark whether it is a special holiday or not;

[0078] c6. Construct a daily traffic flow parameter prediction model;

[0079] c7. Predict daily traffic flow parameters.

[0080] The process of step c1 includes:

[0081] The traffic flow parameters are flow rate and speed, and the data are N consecutive days of data for two years or more; if the speed is to be predicted, only the speed data can be collected, and if the flow rate is to be predicted, both the flow rate and speed data need to be collected;

[0082] The method for collecting traffic flow parameter data is to record the vehicle flow or average speed at the data collection location at a certain time interval, and the time interval can be 5-30 minutes.

[0083] The process of step c2 includes:

[0084] The environmental factors include: day of the week, month, weather;

[0085] The method for marking the environmental factor label is:

[0086] The labeling method of the "day of the week" label is to find the day of the week for each day according to the date of the traffic flow parameter data obtained. If it is Monday, Tuesday to Sunday, it is labeled with 1, 2 to 7 respectively;

[0087] The labeling method of the "month" label is to find out which month each day belongs to in the twelve months of the year according to the date of the traffic flow parameter data obtained. If it belongs to January, February to December, the label is marked with 1, 2 to 12 respectively;

[0088] The labeling method of "weather" is to find the weather conditions of each day according to the date of the acquired traffic flow parameter data. Sunny, cloudy, overcast, light rain, moderate rain, heavy rain, light snow, moderate snow, heavy snow, and others correspond to 1, 2 to 10 respectively. Since the weather is time-varying, it needs to be marked every 2 hours. The marking of each day is carried out in the time sequence of 0:00-2:00, 2:00-4:00, ..., 22:00-24:00, that is, for each day, the label is marked with a one-dimensional vector of length 12. The environmental factor label marking block diagram is shown in the attached figure of the specification. Figure 2 shown.

[0089] The process of step c3 includes:

[0090] The holiday labels include: "Is it a national holiday?" label, "Is it the day before a national holiday?" label, and "Is it the day after a national holiday?" label;

[0091] The method of marking holiday labels is as follows: wherein, the marking method of the label "Whether it is a national holiday or not" is to find out whether each day is a national holiday according to the date of the traffic flow parameter data obtained, and mark the label as 1 or 2 if it is a national holiday or not;

[0092] The labeling method of "Whether it is the day before a national holiday" is to find out whether each day is the day before a national holiday according to the date of the traffic flow parameter data obtained. If it is the day before a national holiday, and if it is not the day before a national holiday, the label is marked with 1 and 2 respectively;

[0093] The labeling method of "Whether it is the day after a national holiday" is to find out whether each day is the day after a national holiday according to the date of the traffic flow parameter data obtained. The day after a national holiday and the day after a non-national holiday are marked with 1 and 2 respectively. The holiday label marking box diagram is shown in the attached figure of the specification. Figure 3 shown.

[0094] In step c4: the daily speed data difference rate S is an indicator for judging the degree of difference between the traffic conditions on any holiday and non-holiday within a given time range, and its value is between 0 and 1. The closer it is to 0, the smaller the difference between the traffic conditions on this holiday and non-holiday, and the closer it is to 1, the greater the difference between the traffic conditions on this holiday and non-holiday.

[0095] In step c5, the daily speed data difference rate S is calculated as follows:

[0096] 1) Filter non-holidays within a given time range to obtain all non-holiday daily speed data, with a total of m time intervals per day; calculate the average speed data of all non-holidays in the same time interval to obtain the average daily speed data data_avg of non-holidays;

[0097] data_avg=[data_avg_1,data_avg_2,...,data_avg_m]

[0098] 2) For any holiday d within a given time range h , whose daily speed data is data_d h ;

[0099] data_d h =[data_1,data_2,...,data_m]

[0100] 3) Calculate the absolute percentage error between the speed data of the holiday and the average daily speed data of non-holidays at each time interval, and obtain the daily speed absolute percentage error vector difference_d h ;

[0101]

[0102] difference_d h =[difference_1,difference_2,...,difference_m]

[0103] Wherein, difference_i is the absolute percentage error of speed in the i-th time interval;

[0104] 4) Set the threshold threshold1;

[0105] 5) Statistic difference_d h The number of difference_i greater than threshold1 is denoted as c, and the daily speed data difference rate S is:

[0106]

[0107] In step c5, the method of marking whether a holiday is special or not is as follows:

[0108] For each holiday within a given time range, the corresponding daily speed data difference rate S is calculated, and a threshold threshold2 is set. If the corresponding S is greater than threshold2, the holiday is considered special and belongs to a special holiday. Otherwise, the holiday is considered not special and belongs to a non-special holiday.

[0109] The labeling method of "Is the holiday special or not" is to label each holiday with 1 for special holidays and 2 for non-special holidays respectively. The labeling diagram of "Is the holiday special or not" is shown in the attached figure of the specification. Figure 4 shown.

[0110] Step c6 is specifically as follows:

[0111] c61. Build a daily traffic flow parameter prediction model that integrates convolutional neural network (CNN) and long short-term memory network (LSTM);

[0112] c62, input data, the first part of the input data is the traffic flow parameter data obtained in step c1, the second part of the input data is the environmental factor label data obtained in step c2, the holiday label data obtained in step c3, and the holiday special or not label data obtained in step c5;

[0113] c63. Rationally divide the acquired traffic flow parameter data and label data into training set and test set;

[0114] c64. Reconstruct the input data of the first part, and reconstruct the traffic flow parameter data into a square matrix with a side length of M*7. M*7 is the time window of the convolution layer. The convolution layer spans the daytime and intraday dimensions. In the daytime dimension, M*7 represents M*7 consecutive days of training data. In the intraday dimension, it represents M*7 time intervals, and M is the optimal week number determined by the input specific traffic flow parameter data; the schematic diagram of the reconstructed traffic flow parameter data is shown in the figure attached to the specification. Figure 5 As shown, t in the figure is the time interval for recording traffic flow parameter data.

[0115] Further explanation of the reconstruction method of input data: If M = 4, and the first day selected is January 1, then the columns of each input element are from the first column of January 1 to the M*7th column of January 28, and the time range included in each column is 28*t. The last row of traffic flow parameter data of the first input element is the traffic flow parameter data of the first time interval t of the corresponding date, the last two rows of traffic flow parameter data of the second input element are the traffic flow parameter data of the first and second time intervals t of the corresponding date, and the traffic flow parameter data of the last input element is the traffic flow parameter data of the last M*7 time intervals t of the corresponding date. The unfilled cells in the matrix are filled with the average traffic flow parameter data of the corresponding date.

[0116] As shown in the attached figure of the instruction manual Figure 6 The figure shows the input diagram of traffic flow parameter data, with a total of m matrices, all of which are inputs of the convolutional layer.

[0117] c65, the convolutional neural network extracts vectors containing daytime and intraday traffic flow parameter information from the reconstructed traffic flow parameter data, and the long short-term memory network performs time series prediction on the output of the convolutional neural network;

[0118] c66. Perform one-hot encoding on the input data of the second part, that is, convert the labeled data of each label into a binary sequence;

[0119] c67. Combine the time series prediction results output by the long short-term memory network with the labeled data of the seven labels converted into one-hot encoding to output the final traffic flow parameter data time series prediction results; the model block diagram is shown in the attached figure of the specification Figure 7 shown.

[0120] c68. Use the test set data to verify the model.

[0121] Step c7 is specifically as follows:

[0122] c71. For each holiday to be predicted, find the "holiday special or not" label of the corresponding historical day, and mark it with the label data with the highest frequency in the historical day labels. If the proportion of 1 and 2 in the historical day labels is the same, then the day is randomly marked as 1 or 2;

[0123] c72. Use the constructed model to predict the daily traffic flow parameters; use the traffic flow parameter data and label data of the previous M*7 days, as well as the label data of the target prediction day, to predict the daily traffic flow parameters of the day. The prediction block diagram is shown in the attached figure of the specification. Figure 8 shown.

[0124] As shown in the attached figure of the instruction manual Fig. 9The following is a comparison chart of the effects of several methods. As can be seen from the figure, the prediction accuracy of the method of the present invention is high, which is better than many existing benchmark methods and shows obvious prediction advantages on some holidays.

Claims

1. A method for predicting daily traffic flow parameters incorporating the impact of holidays, characterized by The method comprises the following steps: c1. Collect traffic flow parameter data; c2. Label environmental factors; c3. Mark holiday labels; the holidays include national holidays, the day before national holidays, and the day after national holidays; c4. Define an indicator that characterizes the degree of difference between traffic conditions on holidays and non-holidays, namely, the daily speed data difference rate S; S is the percentage of time intervals with an absolute percentage error exceeding a certain threshold in the total number of daily time intervals when the daily speed data of any holiday is compared with the average daily speed data of all non-holidays within a given time range, assuming that the traffic flow parameter is selected as speed; c5. Calculate the daily speed data difference rate S and mark whether it is a special holiday or not; c6. Construct a daily traffic flow parameter prediction model; c7. Predict daily traffic flow parameters; Step c6 is specifically as follows: c61. Construct a daily traffic flow parameter prediction model, which integrates convolutional neural network and long short-term memory network; c62, input data, the first part of the input data is the traffic flow parameter data obtained in step c1, the second part of the input data is the environmental factor label data obtained in step c2, the holiday label data obtained in step c3, and the holiday special or not label data obtained in step c5; c63. Rationally divide the acquired traffic flow parameter data and label data into training set and test set; c64. Reconstruct the input data of the first part, and reconstruct the traffic flow parameter data into a square matrix with a side length of M*7. M*7 is the time window of the convolution layer, which spans the daytime and intraday dimensions. In the daytime dimension, M*7 represents M*7 consecutive days of training data, and in the intraday dimension, it represents M*7 time intervals, and M is the optimal week number determined by the input specific traffic flow parameter data; c65, the convolutional neural network extracts vectors containing daytime and intraday traffic flow parameter information from the reconstructed traffic flow parameter data, and the long short-term memory network performs time series prediction on the output of the convolutional neural network; c66. Perform one-hot encoding on the input data of the second part, that is, convert the labeled data of each label into a binary sequence; c67. Combine the time series prediction results output by the long short-term memory network with the labeled data converted into one-hot encoded labels to output the final traffic flow parameter data time series prediction results; c68. Use the test set data to verify the model.

2. The method for predicting daily traffic flow parameters incorporating holiday impacts according to claim 1 is characterized in that: In step c1: The traffic flow parameters are flow rate and speed, and the data are N consecutive days of data for two years or more; if the speed is to be predicted, only the speed data can be collected, and if the flow rate is to be predicted, both the flow rate and speed data need to be collected; The method for collecting traffic flow parameter data is to record the vehicle flow or average speed at the data collection location at a certain time interval.

3. The method for predicting daily traffic flow parameters incorporating holiday impacts according to claim 1 is characterized in that: Step c2 is specifically as follows: The environmental factors include: day of the week, month, weather; The method for marking the environmental factor label is: The labeling method of "day of the week" is to find the day of the week for each day according to the date of the acquired traffic flow parameter data. If it is Monday, Tuesday to Sunday, it is labeled with 1, 2 to 7 respectively; The labeling method of "month" is to find out which month each day belongs to in the twelve months of the year according to the date of the acquired traffic flow parameter data. If it belongs to January, February to December, it is labeled with 1, 2 to 12 respectively; The marking method of the "weather" label is to find the weather conditions of each day according to the date of the traffic flow parameter data that has been obtained. Sunny, cloudy, overcast, light rain, moderate rain, heavy rain, light snow, moderate snow, heavy snow, and other correspond to 1, 2 to 10 respectively. Since the weather is time-varying, it needs to be marked every 2 hours. The marking of each day is carried out in the time sequence of 0:00-2:00, 2:00-4:00, ..., 22:00-24:00, that is, for each day, the label is marked with a one-dimensional vector of length 12.

4. The method for predicting daily traffic flow parameters incorporating holiday impacts according to claim 1 is characterized in that: In step c3: The holiday labels include: "Is it a national holiday?" label, "Is it the day before a national holiday?" label, and "Is it the day after a national holiday?" label; The method for marking holiday labels is: Among them, the labeling method of "Whether it is a national holiday or not" is to find out whether each day is a national holiday according to the date of the acquired traffic flow parameter data day, and mark the label as 1 or 2 respectively if it is a national holiday or not; The labeling method of "Whether it is the day before a national holiday" is to find out whether each day is the day before a national holiday according to the date of the acquired traffic flow parameter data day. If it is the day before a national holiday, and if it is not the day before a national holiday, the label is marked with 1 and 2 respectively; The labeling method of "whether it is the day after a national holiday" is to find out whether each day is the day after a national holiday according to the date of the traffic flow parameter data that has been obtained. The day after a national holiday and the day after a non-national holiday are marked with 1 and 2 respectively.

5. The method for predicting daily traffic flow parameters incorporating holiday impacts according to claim 1 is characterized in that: In step c5, the daily speed data difference rate S is calculated as follows: 1) Filter non-holidays within a given time range to obtain all non-holiday daily speed data, with a total of m time intervals per day; calculate the average speed data of all non-holidays in the same time interval to obtain the average daily speed data data_avg of non-holidays; data_avg=[data_avg_1,data_avg_2,...,data_avg_m] 2) For any holiday d within a given time range h , whose daily speed data is data_d h ; data_d h =[data_1,data_2,...,data_m] 3) Calculate the absolute percentage error between the speed data of the holiday and the average daily speed data of non-holidays at each time interval, and obtain the daily speed absolute percentage error vector difference_d h ; difference_d h =[difference_1,difference_2,...,difference_m] Wherein, difference_i is the absolute percentage error of speed in the i-th time interval; 4) Set the threshold value threshold1; 5) Statistic difference_d h The number of difference_i greater than threshold1 is denoted as c, and the daily speed data difference rate S is:

6. The method for predicting daily traffic flow parameters incorporating holiday impacts according to claim 1 is characterized in that: In step c5, the method of marking whether a holiday is special or not is as follows: For each holiday within a given time range, the corresponding daily speed data difference rate S is calculated, and a threshold threshold2 is set. If the corresponding S is greater than threshold2, the holiday is considered special and belongs to a special holiday. Otherwise, the holiday is considered not special and belongs to a non-special holiday. The labeling method of "Is the holiday special or not" is to mark each holiday with 1 if it is a special holiday and 2 if it is a non-special holiday.

7. The method for predicting daily traffic flow parameters incorporating holiday impacts according to claim 1 is characterized in that: Step c7 is specifically as follows: c71. For each holiday to be predicted, find the "holiday special or not" label of the corresponding historical day, and mark it with the label data with the highest frequency in the historical day labels. If the proportion of 1 and 2 in the historical day labels is the same, then the day is randomly marked as 1 or 2; c72. Use the constructed model to predict daily traffic flow parameters; use the traffic flow parameter data and label marked data of the previous M*7 days, as well as the label marked data of the target prediction day, to predict the daily traffic flow parameters of that day.

Citation Information

Patent Citations

  • Short-time traffic-flow combination prediction method

    CN105046953A

  • Multi-factor short-term traffic flow prediction method based on neural network LSTM

    CN110223510A

  • Short-term traffic flow prediction method based on deep method

    CN110322695A

  • Pedestrian flow prediction device and method

    CN111144648A

  • Business volume prediction method and device, electronic equipment and storage medium

    CN112949886A