Bus arrival time prediction system
By building an LSTM model and combining it with real-time traffic and weather data, the dynamic problem of bus arrival time prediction was solved, efficient and accurate bus arrival time prediction was achieved, and the user experience was improved.
Patent Information
- Application Number
- CN202510852940.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-24
- Publication Date
- 2025-09-12
AI Technical Summary
Traditional bus arrival time prediction methods are unable to dynamically respond to real-time traffic conditions and weather changes, resulting in large prediction errors and affecting passengers' travel arrangements.
Through on-site data collection, preprocessing, feature extraction and machine learning algorithms, an LSTM model is built to predict bus arrival times. Combined with real-time traffic and weather data, the model is optimized to improve accuracy.
It achieves accurate and reliable prediction of bus arrival time, allowing users to understand the riding situation in real time, improving travel efficiency and the reliability of the prediction system.
Smart Images
Figure CN120636191A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of public transportation prediction systems, and in particular to a system for predicting bus arrival times. Background Art
[0002] With the rapid growth of urban populations, traffic congestion is becoming increasingly serious. As a primary mode of public transportation, bus punctuality is directly related to citizens' travel efficiency and quality of life. However, due to factors such as traffic volume, weather, and the times passengers board and alight, bus arrival times are subject to significant uncertainty, making it difficult for passengers to plan their travel schedules. Therefore, accurately predicting bus arrival times has become a pressing technical challenge.
[0003] Traditional bus arrival time prediction methods typically rely on fixed schedules and simple historical data analysis, failing to dynamically adapt to real-time traffic conditions and weather changes, resulting in large prediction errors. In recent years, machine learning algorithms have demonstrated significant potential in traffic forecasting. Training models to analyze and predict complex traffic patterns has begun to be widely used in predicting bus arrival times. However, due to factors such as weather and the number of people boarding and alighting, which can affect bus operation, bus arrival time predictions can be subject to significant errors.
[0004] Therefore, a technical solution is needed to collect and process data on buses, traffic flow, weather, and passenger flow, and use advanced machine learning algorithms to predict bus arrival times, assess the level of congestion based on the number of people on board, and predict stop times to improve the accuracy and reliability of predictions. Summary of the Invention
[0005] To achieve the above objectives, this application provides a bus arrival time prediction system, comprising:
[0006] A field data collection module is used to construct bus driving field data; the bus driving field data is composed of multiple information sets; the multiple information sets include bus trajectory information, station information, traffic flow information and weather information; wherein the bus trajectory information includes bus identification, time point, bus location, traffic flow information includes time point, location, flow status and additional information, wherein the additional information includes: whether it is a platform and the size of the platform; the weather information includes time point and coverage location;
[0007] The data preprocessing module is used to preprocess the bus driving scene data and generate the on-site driving data;
[0008] A feature extraction module is used to call feature engineering to extract features from on-site driving data and construct on-site feature samples; the features include: the current location of the bus, the current driving speed, the current traffic flow, the current weather conditions, the location of the station, and the traffic flow status of the station;
[0009] A time prediction module is used to load a bus arrival time prediction model and predict the time when the bus identifier arrives at the station location based on the on-site feature samples;
[0010] User interface module: used to interact with the user and obtain the user's needs; the user needs include bus identification and bus stop location;
[0011] Result output module: used to obtain user requirements from the user interface module, and according to the user requirements, extract the time for the bus logo associated with the user requirements to arrive at the corresponding bus stop as a prediction result in a specified period, and push it to the user interface module according to the rules;
[0012] Data storage module: used to store bus driving site data to form historical driving data, driving data and prediction results.
[0013] The data preprocessing module preprocesses the bus driving scene data, including:
[0014] Data cleaning, filtering invalid data;
[0015] Using 10 or more adjacent data samples, the K-nearest neighbor algorithm is used for prediction to achieve data completion;
[0016] Standardize weather, vehicle data, GPS, and time information into a unified format;
[0017] Identify abnormal data.
[0018] Feature engineering includes:
[0019] Locate multiple information sets and establish associations between them;
[0020] Extract features from driving data, standardize the features, and form a feature set;
[0021] The feature set is formatted to generate a feature vector containing all features.
[0022] Furthermore, multiple information sets are located including: bus trajectory information, station information, traffic flow information, and weather information;
[0023] Establishing associations between multiple information sets means: associating the bus position of the bus trajectory information with the station position of the station information, associating the bus position and time of the bus trajectory information with the position and time of the traffic flow information, and associating the bus position and time of the bus trajectory information with the coverage position and time point of the weather information.
[0024] Feature engineering extracts historical feature vectors from historical driving data for use in the training environment. When extracting the historical feature vectors, the goal of the prediction model is clarified, and target variables are determined and added from the historical feature vectors. The feature set corresponding to the target variable and the historical feature vector constitutes a complete driving feature set, which is used to train the bus arrival time prediction model.
[0025] The feature engineering extracts field feature vectors from field driving data in a real-time environment for real-time prediction.
[0026] The time prediction module includes a model training unit and a model updating unit;
[0027] The model training unit is used to build a bus arrival time prediction model before loading the bus arrival time prediction model;
[0028] The model updating unit is used to obtain actual arrival time and user feedback to optimize the bus arrival time prediction model.
[0029] Among them, building a bus arrival time prediction model includes:
[0030] The structure of the prediction model is defined as an LSTM model, which includes an input layer, multiple LSTM layers, a fully connected layer, and an output layer: the input layer is used to receive field feature samples;
[0031] The prediction model is trained according to a complete driving feature set, and after the training is completed, a bus arrival time prediction model is generated; wherein the complete driving feature set is composed of a feature set corresponding to a target variable generated based on feature engineering and a historical feature vector.
[0032] Among them, the model update unit includes a feedback submodule for collecting feedback data; the data sources of the feedback submodule are the arrival time predicted by the bus arrival time prediction model, the actual arrival time collected and stored on site, and user feedback information.
[0033] Furthermore, after determining the user's needs, the user interface module initiates a request to the result output module to obtain the prediction results; the user interface module provides a display function for the prediction results, including the display of multiple functions, and supports the display of the real-time location of the bus and the predicted arrival time.
[0034] Furthermore, bus trajectory information also includes vehicle status; vehicle status includes real-time speed, door open / close status, number of passengers getting on and off the bus, and time spent at the platform;
[0035] The platform dwell time is calculated based on the number of passengers getting on and off the bus. The calculation method is:
[0036] T = k × a × max (N, M) + I, where T is the platform dwell time, k is the platform size coefficient, a is the average boarding and alighting time per person, N is the number of boardings, M is the number of alightings, and I is the impact time of vehicles at the same station;
[0037] The impact time of vehicles at the same stop is calculated based on the number of buses at the stop at that time point, combined with the departure difference of buses that have left the station.
[0038] According to the present invention, by collecting and processing information about buses, stations, traffic flow, weather, and the number of people getting on and off, it uses advanced machine learning algorithms to predict congestion levels based on the number of people getting on and off. This information, combined with information about other buses at the same stop and changes in the number of people getting on and off, complements the calculation of arrival times and refines the prediction of bus arrival times, thereby improving the accuracy and reliability of the predictions and enabling users to understand bus ridership in real time. The system also provides feedback to the model based on predicted and actual times, ensuring the validity and reliability of both the data source and the prediction process. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 2. It is a schematic diagram of the structure of a bus arrival time prediction system provided according to an embodiment of the present invention;
[0040] Figure 2 The figure is a data flow diagram of a bus arrival time prediction system provided according to an embodiment of the present invention. DETAILED DESCRIPTION
[0041] This invention provides a bus arrival time prediction system. By collecting and processing information about bus vehicles, traffic flow, weather, and the number of people getting on and off buses, and using advanced machine learning algorithms to assess crowding based on the number of people boarding, this system predicts bus arrival times, improving accuracy and reliability. This system allows users to understand bus ridership in real time. The system encompasses multiple stages, including data collection, preprocessing, feature extraction, model training and optimization, and real-time application, ensuring efficient and accurate processing from data source to prediction results.
[0042] The specific implementation of the present invention is described in detail below with reference to the accompanying drawings.
[0043] The structure of the prediction system provided by the present invention is as follows Figure 1As shown, it includes the following parts:
[0044] P100 field data collection module, used to collect bus driving field data, the bus driving field data consists of multiple information sets;
[0045] Multiple information sets include bus trajectory information, stop information, traffic flow information, and weather information;
[0046] Bus trajectory information includes bus identification, time point, bus location, and vehicle status; the vehicle status includes real-time speed, door open / close status, number of passengers getting on and off, and time spent at the platform;
[0047] 2) Traffic flow information: Traffic flow information includes time point, location, flow status, and additional information. The flow status includes road congestion index, vehicle volume, and average driving speed. Additional information includes whether it is a platform and the size of the platform. When the additional information of the location specified in the traffic flow information is a platform, the information is station information. That is, the location of the station, i.e., the station information, is part of the traffic flow information.
[0048] 3) Weather information includes time point, weather status and coverage location; the weather status includes weather type, rainfall, wind speed information and visibility.
[0049] The data collected in the field data collection module can be obtained by integrating Internet of Things technology. For example, real-time bus trajectory information can be obtained through GPS and bus system interfaces, road condition information can be obtained through traffic monitoring systems, and real-time weather data can be obtained through meteorological stations.
[0050] After collecting the original data, the on-site data collection module also performs preliminary information calculations. Among them, the number of passengers getting on and off the bus during the platform stay time is calculated and then corrected based on the traffic flow information and location information.
[0051] Generally speaking, the platform dwell time is determined by the number of passengers getting on and off the train, which can be expressed as:
[0052] T = k × a × max (N, M) + I, where T is the platform dwell time, k is the platform size coefficient, a is the average boarding and alighting time per person, N is the number of boardings, M is the number of alightings, and I is the impact time of vehicles at the same station;
[0053] When the platform scale is larger, the value of the platform scale coefficient is larger, the time for a single person to get on and off the train is affected by the platform scale coefficient, and the platform stay time is longer;
[0054] The impact time of vehicles at the same stop is calculated by combining the number of buses at the stop at that time point with the departure difference of buses that have left the station.
[0055] Field data collection module realizes data collection such as Figure 2 As shown in the figure, the bus driving scene data is obtained through the timing task and GPS equipment, traffic flow API, weather API interface, and on-board passenger flow equipment (such as on-board camera), and Figure 1 As shown in , the subsequent transmission direction of the data is defined according to the needs, including transmission to the P160 data storage module for storage to form the bus travel history data, and transmission to the P110 data preprocessing module for preprocessing to form travel data, ready for data processing by the machine learning model, combined with the station information, and predicting the time to each station.
[0056] The field data collection module has certain requirements for the real-time performance of data collection and needs to obtain prediction results based on the latest field data.
[0057] P110 data preprocessing module is used to preprocess the bus driving scene data and generate on-site driving data;
[0058] The data preprocessing module performs preprocessing operations on the collected data, including data cleaning, data completion, standardization, and removal of outliers to ensure data quality and consistency.
[0059] 1) Data Cleaning: This process filters out invalid data, for example, by using ranges to remove outliers in GPS coordinates and weather values. Specifically, we use GPS devices and APIs to convert data into a format suitable for Spark computations. We predefine reasonable value ranges (for example, visibility data should be within a range of 0 to 200 meters). This data is then encapsulated in the Spark DataFrame filter method, combined with comparison operators for data cleaning and standardization.
[0060] 2) Data completion: If weather conditions or vehicle speeds are missing, data completion is achieved by using 10 or more adjacent data samples and using the K-nearest neighbor algorithm for prediction.
[0061] 3) Standardization: Manage weather, vehicle data, GPS, time and other information in a unified format, as shown in Tables 1 to 3:
[0062] Table 1: Traffic flow information
[0063]
[0064] Table 2: Weather data
[0065]
[0066]
[0067] Table 3: Bus trajectory information
[0068]
[0069] 4) Abnormal Data: After acquiring real-time data, Spark Streaming is used to set up sliding windows, defining the window size and sliding interval. For each data point within the sliding window, statistical metrics such as the mean and standard deviation are calculated. The mean is calculated by dividing the sum of all data points within the window by the number of data points. The standard deviation is calculated based on the variance (the variance is the sum of the squares of the difference between each data point and the mean, divided by the number of data points; the standard deviation is the square root of the variance). For each data point within the window, a Z-score is calculated (the Z-score is the difference between the data point and the mean divided by the standard deviation). If the absolute value of the Z-score is greater than the set threshold, the data point is considered an anomaly.
[0070] The P120 feature extraction module is used to call feature engineering to extract features from the on-site driving data processed by the P110 data preprocessing module and construct on-site feature samples; the features include: the current location of the bus, current driving speed, current traffic flow, current weather conditions, station location and station flow status, etc.
[0071] The feature engineering includes:
[0072] 1) Locate multiple information sets and establish associations between them;
[0073] The multiple information sets located in this step are consistent with the information sets collected in the P100 field data collection module, including bus trajectory information, traffic flow information and weather information, where the traffic flow information includes station information.
[0074] Establishing associations between multiple information sets means: associating the bus position of the bus trajectory information with the station position of the station information, associating the bus position and time of the bus trajectory information with the position and time of the traffic flow information, and associating the bus position and time of the bus trajectory information with the coverage position and time point of the weather information.
[0075] Meanwhile, the bus track information, traffic flow information and weather information include a field information set and a historical information set, and the historical information set is data from the data storage module.
[0076] 2) Extract features from driving data, standardize the features, and form a feature set;
[0077] For example, use Spark to create a SparkSession to read a file storing bus location data, and then use Scala to create a SparkSession instance, which can be converted into a (String, Double, Double) type data set for subsequent calculations.
[0078] This step also involves merging the data from the associated information sets into a complete dataset. For example, we can merge the station information with the GPS coordinates of the weather data's overlay location, the GPS coordinates of the station location information, and the timestamp of the bus trajectory information to obtain a dataset of the station information and weather conditions that the bus passed during a specified time period. To perform the merge, if using Python, use the DataFrame join method, for example, current_location_df.join(speed_df,on="bus_id") to perform the merge operation.
[0079] 3) Formatting the feature set to generate a feature vector containing all features.
[0080] Feature engineering can extract historical feature vectors from historical driving data for use in the training environment. When extracting historical feature vectors, the goal of the prediction model is clarified, and the target variable is determined and added from the historical feature vectors. For example, when predicting bus delays, the target variable is a binary variable (0 means no delay, 1 means delay). Assuming that the structure of the delay dataset includes bus ID, station location, time, and delay status, after reading the dataset, it is associated with the historical driving data through the bus ID, and the date and delay status columns are added as target variables to the feature dataset, and then added through the join operation.
[0081] Predict bus arrival times at all stops. The target variable is a two-dimensional array [stop location, actual arrival time]. The feature set corresponding to the target variable and the historical feature vectors constitutes the complete travel feature set. This complete travel feature set can be used to train a bus arrival time prediction model.
[0082] Feature engineering can extract on-site feature vectors from on-site driving data in a real-time environment for real-time prediction. After inputting the on-site feature vectors into the bus arrival time prediction model, the predicted bus arrival time can be obtained.
[0083] like Figure 1 As shown in , the field feature samples generated in this step are accumulated by the P160 data storage module and used for model training and updating; they can also be used to predict arrival time in the time prediction module in combination with user requirements from the user interface module obtained through the result output module.
[0084] P130 time prediction module, used to load the bus arrival time prediction model and predict the time when the bus logo arrives at the location of all bus stops based on the on-site feature samples;
[0085] The time prediction module includes P131 model training unit and P132 model update unit;
[0086] The P131 model training unit is used to build a bus arrival time prediction model before loading the bus arrival time prediction model;
[0087] When constructing a bus arrival time prediction model, the application of decision trees, support vector machines and neural networks is combined, which plays an important role in the application scenario of the present invention.
[0088] Specifically, building a bus arrival time prediction model includes:
[0089] The structure of the prediction model is defined as an LSTM model, which includes an input layer, multiple LSTM layers, a fully connected layer, and an output layer: the input layer is used to receive on-site feature samples; the LSTM layer is used to learn long-term dependencies in time series, and its internal cell state and gating mechanism can effectively process and transmit time series information; multiple LSTM layers can further extract complex features of the data; the fully connected layer is used to map the features output by the LSTM layer to the final prediction dimension, and the output layer is used to output the predicted value of the bus arrival time.
[0090] In this step, the prediction model is trained based on the complete driving feature set. After the training is completed, a bus arrival time prediction model is generated; wherein the complete driving feature set is composed of a feature set corresponding to the target variable generated based on feature engineering and the historical feature vector.
[0091] When training the model, the prepared sample data is split into a training set and a test set, which are then fed into the prediction model. For LSTM models, the sample data must be organized in a time series format, typically in batches. For example, each batch contains sample data from multiple time steps, with each time step containing all features. During training, the model continuously adjusts internal parameters (such as the weights of the LSTM layer) to learn the relationship between sample features and the target value (bus arrival time). A loss function is defined to measure the difference between the model's prediction and the actual target value, such as mean squared error (MSE) for regression problems. An optimizer (such as stochastic gradient descent (SGD) or Adam) is used to minimize the loss function. The optimizer uses a backpropagation algorithm to calculate the gradient of the loss function with respect to the model parameters and updates the parameters based on the gradient, gradually reducing the model's prediction error on the training data. Once the target state is reached, the bus arrival time prediction model is trained and generated. After training, the test set is used to evaluate model performance to ensure accuracy and generalization.
[0092] The bus arrival time prediction model generated in this step is deployed to the prediction system for loading and use, so as to receive the field feature vector generated in step P120 in real time. The bus arrival time prediction model can predict the arrival time of each station through the field feature vector. The prediction results are directly related to the current traffic conditions and weather changes. For example, if the real-time traffic flow shows road congestion, the bus arrival time in the prediction results will be delayed accordingly. This is because during the training process, the bus arrival time prediction model has learned the relationship between traffic congestion and bus arrival time delays; similarly, for weather changes, such as severe weather (heavy rain, heavy snow) that may cause the bus to slow down, the bus arrival time prediction model will also adjust the prediction results based on the relationship between weather conditions and travel time in the training data, thereby improving the timeliness and reliability of the prediction.
[0093] The P132 model updating unit is used to obtain actual arrival time and user feedback to optimize the bus arrival time prediction model.
[0094] The model update unit includes a feedback submodule for collecting feedback data; Figure 2 As shown in Figure 1, the arrival time predicted by the bus arrival time prediction model, the actual arrival time collected and stored on-site, and the user's feedback information constitute the data source of the feedback submodule;
[0095] By using the feature data provided by the feedback submodule, the bus arrival time prediction model can achieve incremental learning and online learning, enabling the model to adapt to the ever-changing traffic environment and improve long-term prediction results.
[0096] P140 user interface module: used to interact with the user and obtain the user's needs; the user needs include bus identification and bus stop location;
[0097] The user interface module uses the Uniapp framework to build the user end, supports cross-platform operation, and implements location tracking and real-time updates through uni.getLocation(OBJECT) and wx.onLocationChange;
[0098] The user interface supports the needs of ordinary users, including providing and collecting user demand information, such as bus signs and platform signs to form specific information of user needs;
[0099] In addition to supporting the functions required by administrator roles, the user interface also provides a management interface. The management interface uses the Vue.js framework and provides data monitoring, system configuration, and user management functions. It allows administrators to view and manage the system's operating status.
[0100] After determining the user's needs, the user interface module initiates a request to the P150 result output module to obtain the prediction results; the user interface module provides display functions for the prediction results, including the display of user-defined reminders and offline queries, and supports the display of the bus's real-time location and predicted arrival time.
[0101] P150 result output module: used to obtain user needs from the user interface module, and based on the user needs, extract the time for the bus logo associated with the user needs to arrive at the corresponding bus stop as a prediction result according to the specified period, and push it to the user interface module according to the rules.
[0102] The result output module can realize the back-end functions of the management end: it adopts the SpringBoot framework, provides API interfaces for the user interface module to call, realizes system configuration, data management and user management functions, and ensures the stable operation and efficient maintenance of the system.
[0103] P160 data storage module: used to store bus driving site data, driving data and prediction results.
[0104] The data storage module uses MySQL and NoSQL (such as MongoDB) databases to store data. MySQL is used for structured data and complex query scenarios, while NoSQL is suitable for large data volumes and flexible data model scenarios.
[0105] Based on the prediction system provided by the present invention, it is possible to predict the arrival time of buses at various bus stops. Based on the accuracy, reliability, and timeliness of the prediction results, it is possible to further provide extended functions, such as providing suggestions for the best bus route, departure time, and estimated arrival time based on the departure and destination, thereby realizing a route planning function; supporting user-defined arrival reminders in terms of data, promptly notifying users of the boarding time, and also enabling the function of querying bus arrival time and route information in an offline environment. In the case of an integrated navigation function, it can help users navigate to their destination in real time. At the same time, the functions of the present invention can also be integrated with services of third-party service providers (such as online car-hailing, food delivery, etc.), providing more travel options and convenience.
[0106] In summary, this invention provides a bus arrival time prediction system based on machine learning algorithms and real-time data processing. By collecting and processing data on bus vehicles, traffic flow, weather, and other data, and utilizing advanced machine learning algorithms to predict bus arrival times, the system improves accuracy and reliability. The implementation of the prediction system also comprehensively considers security, privacy protection, scalability, and maintainability requirements, providing citizens with an accurate and convenient bus travel experience.
[0107] The above disclosures are only a few specific embodiments of the present invention. However, the present invention is not limited thereto. Any changes that can be conceived by those skilled in the art should fall within the scope of protection of the present invention.
Claims
1. A bus arrival time prediction system, characterized in that: include: A field data collection module is used to construct bus driving field data, the bus driving field data being composed of multiple information sets; the multiple information sets including bus trajectory information, station information, traffic flow information, and weather information; wherein the bus trajectory information includes bus identification, time point, and bus location; the traffic flow information includes time point, location, flow status, and additional information, wherein the additional information includes whether it is a platform and the size of the platform; and the weather information includes time point and coverage location; The data preprocessing module is used to preprocess the bus driving scene data and generate the on-site driving data; A feature extraction module is used to call feature engineering to extract features from the on-site driving data and construct on-site feature samples; the features include: the current location of the bus, the current driving speed, the current traffic flow, the current weather conditions, the location of the station and the traffic flow status of the station; A time prediction module is used to load a bus arrival time prediction model and predict the time when the bus identifier arrives at the station location based on the on-site feature samples; User interface module: used to interact with the user and obtain the user's needs; the user needs include bus identification and bus stop location; A result output module is used to obtain user requirements from the user interface module, and extract the time for the bus logo associated with the user requirements to arrive at the corresponding bus stop as a prediction result according to the user requirements in a specified period, and push it to the user interface module according to the rules; Data storage module: used to store bus driving site data to form historical driving data, driving data and prediction results.
2. The prediction system according to claim 1, characterized in that The data preprocessing module preprocesses the bus driving scene data, including: Data cleaning, filtering invalid data; Using 10 or more adjacent data samples, the K-nearest neighbor algorithm is used for prediction to achieve data completion; Standardize weather, vehicle data, GPS, and time information into a unified format; Identify abnormal data.
3. The prediction system according to claim 1, wherein: The feature engineering includes: Locate multiple information sets and establish associations between them; Extracting features from the driving data, and standardizing the features to form a feature set; The feature set is formatted to generate a feature vector containing all features.
4. The prediction system according to claim 3, characterized in that The plurality of positioning information sets include: bus track information, station information, traffic flow information and weather information; The establishing association between multiple information sets refers to: associating the bus position of the bus trajectory information with the station position of the station information, associating the bus position and time of the bus trajectory information with the position and time of the traffic flow information, and associating the bus position and time of the bus trajectory information with the coverage position and time point of the weather information.
5. The prediction system according to claim 4, characterized in that The feature engineering extracts historical feature vectors from historical driving data for use in the training environment. When extracting the historical feature vectors, the goal of the prediction model is clarified, and target variables are determined and added from the historical feature vectors. The feature set corresponding to the target variable and the historical feature vector constitutes a complete driving feature set, which is used to train the bus arrival time prediction model. The feature engineering extracts field feature vectors from field driving data in a real-time environment for real-time prediction.
6. The prediction system according to claim 5, characterized in that The time prediction module includes a model training unit and a model updating unit; The model training unit is used to build a bus arrival time prediction model before loading the bus arrival time prediction model; The model updating unit is used to obtain actual arrival time and user feedback, and optimize the bus arrival time prediction model.
7. The prediction system according to claim 6, characterized in that The bus arrival time prediction model is constructed, including: The structure of the prediction model is defined as an LSTM model, which includes an input layer, multiple LSTM layers, a fully connected layer, and an output layer: the input layer is used to receive field feature samples; The prediction model is trained according to a complete driving feature set, and after the training is completed, a bus arrival time prediction model is generated; wherein the complete driving feature set is composed of a feature set corresponding to a target variable generated based on feature engineering and a historical feature vector.
8. The prediction system according to claim 6, characterized in that The model updating unit includes a feedback submodule for collecting feedback data; the data sources of the feedback submodule are the arrival time predicted by the bus arrival time prediction model, the actual arrival time collected and stored on site, and user feedback information.
9. The prediction system according to claim 1, wherein: After determining the user's needs, the user interface module initiates a request to the result output module to obtain the prediction results; the user interface module provides a display function for the prediction results, including a display of functions, and supports the display of the real-time location of the bus and the predicted arrival time.
10. The prediction system according to claim 1, wherein: The bus trajectory information also includes vehicle status; the vehicle status includes real-time speed, door switch status, number of passengers getting on and off the bus, and platform stay time; The platform dwell time is calculated based on the number of passengers getting on and off the bus. The calculation method is: T = k × a × max (N, M) + I, where T is the platform dwell time, k is the platform size coefficient, a is the average boarding and alighting time per person, N is the number of boardings, M is the number of alightings, and I is the impact time of vehicles at the same station; The impact time of vehicles at the same stop is calculated based on the number of buses at the stop at that time point, combined with the departure difference of buses that have left the station.