Robust prediction method for bus arrival time based on sparse GPS data with missing arrival information

Abstract

The present invention relates to the field of smart transportation technology, and specifically to a robust bus arrival time prediction method for sparse GPS data with missing arrival information. The method comprises generating a representative stop geohash grid using an improved clustering algorithm based on historical vehicle GPS data and bus stop locations to determine whether a vehicle has arrived at a stop. A multi-scale spatiotemporal network diagram is constructed using data on bus routes, urban points of interest, and external environments, with nodes representing geohash grids of different scales and edge attributes representing vehicle travel time characteristics at different times and spaces. Finally, a machine learning algorithm is used to fuse real-time GPS data with network diagram attributes to predict bus arrival times. Experimental verification shows that in a test set with a 30% data downsampling, the prediction error is reduced by 22% compared to traditional methods, improving the robustness of arrival time prediction in low-data-quality environments.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and more specifically, to a robust prediction method for bus arrival time based on sparse GPS data with missing arrival information. Background Art

[0002] Although public transportation networks cover vast swaths of the city, seamless connectivity across the entire region remains a challenge. For areas beyond the reach of urban rail transit, such as suburban areas and areas between rail lines, public bus routes play a crucial role in meeting the basic travel needs of the general public. Providing accurate bus arrival time prediction services can optimize on-time bus routes, enhance refined management, help passengers plan their travel times, and improve the convenience of public transportation.

[0003] Public transportation data reported by different cities varies. While Beijing's bus system requires passengers to swipe their bus card before boarding and exiting, it can capture arrival data. Shanghai, however, only requires swiping their card before boarding, resulting in missing arrival information. GPS packet loss rates exceed 40% in complex environments like tunnels. Conventional arrival time prediction models lack the ability to fully understand the complex environments in which vehicles operate and the geospatial network they exploit, making it difficult to accurately predict arrival times in such low-data-quality environments.

[0004] Therefore, it is urgent to build a method that can ensure the accuracy of bus arrival time prediction in scenarios with missing information or missing GPS data, and improve the robustness of bus arrival time prediction in low data quality environments. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a robust prediction method for bus arrival time based on sparse GPS data with missing arrival information.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A robust prediction method for bus arrival time based on sparse GPS data with missing arrival information is proposed. The specific process is as follows:

[0008] Step 1: Cluster to obtain the geo-hash network of common stops.

[0009] like Figure 2 As shown in the figure, the K-means clustering algorithm is used to cluster the historical GPS data of vehicles near the bus station, and the geographical hash grid groups of vehicles stopping at the bus station on different bus lines are identified;

[0010] Step 2: Construct a geospatial bus driving network map.

[0011] Based on vehicle GPS data, bus stop location latitude and longitude, and city Point of Interest (POI) latitude and longitude, these data are encoded into geohash grids of varying precision to construct a multi-scale geospatial bus network map, allowing different bus routes to share the same geohash grid. Even if vehicle GPS data is sparse or missing, a series of geohash grids can be matched from the network map in the direction of the bus stop using the vehicle's current location.

[0012] Step 3: Construct the attributes of the relationship edges in the network graph.

[0013] By using the historical GPS data of vehicles in different time and space, the travel time between bus stops is constructed as the attribute of the network graph edge;

[0014] Step 4: Train the model to predict arrival time.

[0015] Based on historical vehicle GPS data, arrival time data is constructed. Using the nodes and edges of the constructed multi-scale spatiotemporal network, a model is trained to learn the historical driving patterns of vehicles and predict the arrival time of bus routes in real time.

[0016] Furthermore, the specific process of step one is as follows:

[0017] The pigeon flock optimization algorithm is used to optimize the initial cluster centers of K-means clustering, improve the clustering effect, and enhance the representativeness of the selected geohash grids;

[0018] Apply statistical significance analysis method, perform variance analysis on each grid, and select the geographic hash grid that can represent the characteristics of the group;

[0019] The representativeness of the grid is evaluated through cross-validation, and the historical vehicle GPS data within 300 meters from the bus stop are screened as the candidate set of geo-hash grids for selecting the stop sites, which are used as the basis for judging the arrival of buses.

[0020] Furthermore, step 2 constructs a geospatial bus driving network diagram, and the specific process includes:

[0021] Data collection: Collect bus GPS trajectory data, including timestamp, latitude and longitude, speed, and direction; also collect bus stop latitude and longitude and POI latitude and longitude;

[0022] Grid encoding: Use the Geohash algorithm to encode the longitude and latitude into a string of length n to form geohash grids of different sizes, where n is the grid precision and the value range is ;

[0023] Multi-scale grid generation: Generate multi-scale grid systems, such as 5-bit, 6-bit, and 7-bit precision grids, to form a set of geohash grids with different spatial resolutions;

[0024] Trajectory mapping: Map the GPS data reported by the vehicle's driving trajectory to a geohash grid of corresponding precision (Geohash string with different digits) in time series to form a trajectory point sequence:

[0025] ;

[0026] in, ;

[0027] Spatiotemporal network graph construction: Construct a spatiotemporal network graph, where nodes are all the geo-hash grids that appear, including GPS track networks, bus stop grids, and POI grids. If the number of consecutive appearances of a vehicle GPS track grid in the time series exceeds a threshold, a spatiotemporal relationship edge is formed. If the vehicle GPS track grid is the same as the bus stop or POI grid, a matching relationship edge is formed.

[0028] Furthermore, in step 3, the attributes of the relationship edges of the network graph are constructed. The specific process includes:

[0029] Driving time calculation: Based on the adjacent geohash grid pairs in the vehicle's historical GPS trajectory sequence, extract their timestamps and calculate the driving time , the specific calculation formula is as follows:

[0030] ;

[0031] in, Vehicle GPS data at all times and The vehicle GPS data at a given moment is the two preceding and following data in the GPS trajectory sequence;

[0032] Statistical analysis: For grids of different scales, the mean of travel time is calculated. and standard deviation , the specific calculation formula is as follows:

[0033] ;

[0034] in, is the data length, is the driving time;

[0035] Acquisition of spatiotemporal features: Obtain historical weather information by geo-hash grid division; divide morning peak hours (07:00-09:00) and evening peak hours (17:00-19:00) by timestamp; mark dates as weekdays, weekends, or statutory holidays;

[0036] Attribute storage: A sliding window mechanism is used to calculate the mean and standard deviation of influencing factors such as weather, morning and evening rush hours, and holidays, and store them as edge attributes.

[0037] Furthermore, in step 4, the model is used to predict the arrival time of buses. The specific process is as follows: Figure 1 As shown:

[0038] When making predictions, the system first matches nodes in a pre-built multi-scale spatiotemporal network diagram based on the real-time GPS data obtained by the vehicle and the GPS data of the target bus stop to be predicted. After accurately locating the corresponding nodes, the model extracts the attribute information of these nodes themselves, as well as the attribute data of the relationship edges between nodes, and finally outputs the predicted arrival time value of the target bus stop to be predicted, realizing effective prediction of the real-time arrival time of buses in low-data quality environments.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. Data-driven innovation: No need to rely on additional hardware equipment, only using existing GPS data, bus stop locations and POI data, reducing costs and increasing applicability;

[0041] 2. Algorithm fusion advantages: The clustering process is optimized by combining the pigeon flock optimization algorithm to improve the accuracy of station grid judgment; the multi-scale space-time network diagram fully considers the spatial and temporal dimensions to enhance the expression of data features;

[0042] 3. Accurate prediction: Through multi-source data feature extraction and machine learning model training, the accuracy of bus arrival time prediction in low-data quality environments is effectively improved, providing reliable support for traffic scheduling and public travel. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of a robust bus arrival time prediction method for sparse GPS data with missing arrival information;

[0044] Figure 2 This is a schematic diagram of the present invention for determining whether a vehicle has arrived at a station based on vehicle GPS data. DETAILED DESCRIPTION

[0045] Example 1, refer to Figure 1 As shown, the robust prediction method for bus arrival time with sparse GPS data with missing arrival information in this embodiment has the following process:

[0046] Step 1: Cluster to obtain the geographical hash grid of common stops.

[0047] When passengers get on and off a bus at a stop, the vehicle's position cannot be moved. However, Shanghai buses do not have devices to collect data on the opening and closing of doors when passengers get on and off, nor do they have data on users swiping their cards when they get off the bus. In addition, there is sparse delay in the GPS data reported by the vehicle. This low data quality environment makes it very difficult to accurately determine whether the bus has arrived at the stop.

[0048] like Figure 2 As shown in the figure: K-means clustering algorithm is used to cluster the historical GPS data of vehicles near bus stops and identify the geographical hash grid groups where vehicles of different bus lines stop at the stops. The specific method is as follows:

[0049] 1. Use the pigeon flock optimization algorithm to optimize the initial cluster centers of K-means clustering, improve the clustering effect, and enhance the representativeness of the selected geohash grids. The specific process includes:

[0050] 1.1. Data preprocessing: Input data is the feature vector of the geohash grid , the target variable is to optimize the initial cluster center of K-means , so that the initial centers are distributed in the core areas of different data clusters as much as possible, and each initial center candidate is decoded into a dimension of The vector ( is the feature dimension);

[0051] 1.2. Initialize the pigeon population: randomly select from the dataset samples as the initial candidate centers to ensure that each individual Each center is different from the others;

[0052] 1.3. Global Exploration: The K-means intra-cluster sum of squares is used as the optimization target to measure the clustering quality. The specific calculation formula is:

[0053] ;

[0054] in, Is assigned to The goal is to minimize the sample set of clusters For each individual , based on the current location , global optimal position and its own historical best position Update the position. The specific calculation formula is as follows:

[0055] ;

[0056] in for Random numbers within is the number of iterations;

[0057] 1.4. Local development: Perform a detailed search near the global optimal area. Each individual adjusts its position based on its own motion information. The specific calculation formula is as follows:

[0058] ;

[0059] in, is the scaling factor, is the mean vector of the individual's current position.

[0060] 1.5. Iterative optimization: Repeat steps 3-4 until the change rate of the sum of squares within the cluster is less than 1% for 5 consecutive generations, and output the global optimal individual As the initial cluster center of K-means;

[0061] 2. Apply statistical significance analysis to perform variance analysis on each grid. The difference between the eigenvalues ​​of each grid and those of other groups is calculated. The geohash grids that represent the characteristics of this group are selected. The selection criteria include proximity to bus stops, high GPS point density, and avoiding redundant GPS data reported before and after bus arrival.

[0062] 3. Grid representativeness was evaluated through cross-validation. Historical vehicle GPS data within 300 meters of bus stops was selected as a candidate set of geohash grids for selecting bus stops. The selected stop grids matched more than 90% of the historical vehicle arrival GPS data. This was used as the basis for determining bus arrivals, minimizing the actual bus parking locations and the passengers' boarding and alighting experiences.

[0063] Step 2: Construct a geospatial bus driving network map.

[0064] Based on the vehicle's GPS data, the longitude and latitude of the bus stop location, and the longitude and latitude of the city's points of interest, it is encoded into geo-hash grids of varying precision to construct a multi-scale geospatial bus vehicle travel network diagram, allowing different bus routes to share the same geo-hash grid. Even if the vehicle's GPS data is sparse or missing, a series of geo-hash grids can be matched from the network diagram according to the direction of the bus stop based on the vehicle's current location. The specific method is as follows:

[0065] Data collection: Collect bus GPS trajectory data, including timestamp, latitude and longitude, speed, and direction; also collect bus stop latitude and longitude and POI latitude and longitude;

[0066] Grid encoding: Use the Geohash algorithm to encode the longitude and latitude into a string of length n to form geohash grids of different sizes, where n is the grid precision and the value range is ;

[0067] Multi-scale grid generation: Generate multi-scale grid systems, such as 5-bit, 6-bit, and 7-bit precision grids, to form a set of geohash grids with different spatial resolutions;

[0068] Trajectory mapping: Map the GPS data reported by the vehicle's driving trajectory to a geohash grid of corresponding precision (Geohash string with different digits) in time series to form a trajectory point sequence:

[0069] ;

[0070] in, ;

[0071] Spatiotemporal network graph construction: Construct a spatiotemporal network graph, where nodes are all the geo-hash grids that appear, including GPS track networks, bus stop grids, and POI grids. If the number of consecutive appearances of a vehicle GPS track grid in the time series exceeds a threshold, a spatiotemporal relationship edge is formed. If the vehicle GPS track grid is the same as the bus stop or POI grid, a matching relationship edge is formed.

[0072] Step 3: Construct the attributes of the relationship edges in the network graph.

[0073] By using the historical GPS data of vehicles in different time and space, the driving time between bus stops is constructed as the attribute of the network graph edge. The details are as follows:

[0074] Driving time calculation: Based on the adjacent geohash grid pairs in the vehicle's historical GPS trajectory sequence, extract their timestamps and calculate the driving time , the specific calculation formula is as follows:

[0075] ;

[0076] in, Vehicle GPS data at all times and The vehicle GPS data at a given moment is the two preceding and following data in the GPS trajectory sequence;

[0077] Statistical analysis: For grids of different scales, the mean of travel time is calculated. and standard deviation , the specific calculation formula is as follows:

[0078] ;

[0079] in, is the data length, is the driving time;

[0080] Acquisition of spatiotemporal features: Obtain historical weather information by geo-hash grid division; divide morning peak hours (07:00-09:00) and evening peak hours (17:00-19:00) by timestamp; mark dates as weekdays, weekends, or statutory holidays;

[0081] Attribute storage: A sliding window mechanism is used to calculate the mean and standard deviation of influencing factors such as weather, morning and evening rush hours, and holidays, and store them as edge attributes.

[0082] Step 4: Train the model to predict arrival time.

[0083] Based on historical vehicle GPS data, we construct arrival time data. We use the nodes and edges of the constructed multi-scale spatiotemporal network to train a model to learn the historical driving patterns of vehicles and predict the arrival time of bus routes in real time. The specific steps are as follows:

[0084] Training data construction: Using historical vehicle GPS data time series, according to step 1, filter the GPS data that is less than 300 meters away from the target bus stop and hits the bus stop geo-hash grid, calculate the time of arrival at the target bus stop, and use the remaining GPS data as the input of the training data set. The goal is to predict the arrival time ;

[0085] Model construction: The Xgboost model was selected, and the selection of various parameters was based on balancing model performance, avoiding overfitting, and ensuring computational efficiency. The maximum depth max_depth was set to 8, ensuring that the model captured the complex characteristics of the data while preventing overfitting due to excessive tree depth; learning_rate was set to 0.1, so that the model could converge quickly during training and not easily skip the optimal solution; n_estimators was set to 500, taking into account both prediction accuracy and computing resources; reg:absoluteerror was selected as the loss function and mae as the evaluation function, which is more robust to outliers that may appear in the prediction of bus arrival time; subsample was set to 0.8, and each tree was trained by randomly sampling 80% of the samples, which effectively reduced the model variance, simulated the sparse GPS data, and improved the model generalization ability;

[0086] Model training: Use and The node attributes of the corresponding geographic hash grid, such as id encoding, scale, administrative region, etc.; the attributes of the relationship edge, such as the mean and standard deviation of historical driving time, are used as the input of the model, and the output of the model is , the Mean Absolute Error (MAE) is used to evaluate the performance of the model. The specific calculation method is as follows:

[0087] ;

[0088] Where m is the dataset size, Is it really the arrival time? To predict the arrival time;

[0089] Real-time prediction: Figure 1 As shown in the figure, during the prediction, the nodes in the multi-scale spatiotemporal network graph are matched according to the real-time GPS data of the vehicle and the GPS data of the target bus stop to be predicted, and the attribute data of the nodes and relationship edges are obtained as the input of the model to obtain the predicted arrival time of the target bus stop to be predicted.

[0090] The real-time bus arrival time prediction solution provided by the present invention uses a clustering algorithm combined with an optimization algorithm to obtain a geographical hash grid of common vehicle parking locations, which is used to determine whether the vehicle has arrived at the bus stop. This makes up for the lack of vehicle arrival information data in the original reported data in the application scenario of the present invention. In response to the situation where vehicle GPS data reporting is sparse or missing, the solution fully utilizes the vehicle's historical GPS data and urban point of interest data to construct the nodes and relationship edges of a multi-scale geographic spatial network diagram. In combination with influencing factors such as dynamic weather and holidays, the accuracy of bus arrival time prediction in low-data-quality environments is effectively improved.

[0091] An experimental comparison was conducted using a historical vehicle GPS dataset from three bus routes in Shanghai, using the same decision tree model constructed using a traditional method and the proposed method. Simulating a real-world scenario with poor data quality, the proposed method performed on a 30% downsampled test set. The mean absolute error of the proposed method decreased by approximately 22%, demonstrating the feasibility of the proposed method in practical applications.

[0092] The above formulas are all dimensionless and numerically calculated, and the preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0093] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0094] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0095] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0096] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0097] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0098] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0099] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A robust prediction method for bus arrival time with sparse GPS data with missing arrival information, characterized by: The method flow is as follows: Step 1: Clustering to obtain the geo-hash network of common stops: Use the K-means clustering algorithm to identify the geo-hash grid groups where vehicles on different bus routes stop at the stops; Step 2: Construct a geospatial bus driving network diagram: collect bus operation data, encode and generate a multi-scale geohash grid, map GPS trajectories to the grid to form a sequence, and construct a spatiotemporal network diagram; Step 3: Construct the attributes of the network graph edges: Use the historical GPS data of buses to construct the driving time between bus stops as the attributes of the network graph edges; Step 4: Train the model to predict bus arrival times: Build arrival time data based on historical vehicle GPS data. Use a multi-scale spatiotemporal network node and edge training model to learn historical driving patterns and predict bus arrival times in real time.

2. The robust prediction method for bus arrival time based on sparse GPS data with missing arrival information according to claim 1 is characterized in that: The specific process of step one includes: The pigeon flock optimization algorithm is used to optimize the initial cluster centers of K-means clustering; Apply statistical significance analysis method, perform variance analysis on each grid, and select the geographic hash grid that can represent the characteristics of the group; The representativeness of the grid is evaluated through cross-validation, and the historical vehicle GPS data within 300 meters from the bus stop are screened as the candidate set of geo-hash grids for selecting the stop sites, which are used as the basis for judging the arrival of buses.

3. The robust prediction method for bus arrival time based on sparse GPS data with missing arrival information according to claim 2 is characterized in that: The pigeon flock optimization algorithm is used to optimize the initial cluster center of K-means clustering. The specific process includes: Data preprocessing: Input data is the feature vector of the geohash grid , the goal is to optimize the initial clustering center of K-means , decode each initial center candidate into a dimension of vector, is the feature dimension; Initialize the pigeon population: construct the initial candidate center; Global exploration: The K-means cluster sum of squares is used as the optimization target; for each individual, the position is updated based on the current position, the global optimal position, and the individual's historical optimal position; Local development: Perform a detailed search near the global optimal area, and each individual adjusts its position based on its own motion information; Iterative optimization: Repeat global exploration and local development until the change rate of the sum of squares within the cluster is less than 1% for five consecutive generations, and output the global optimal individual as the initial cluster center of K-means.

4. The robust prediction method for bus arrival time based on sparse GPS data with missing arrival information according to claim 1 is characterized in that: Step 2: Construct a geospatial bus network diagram. The specific process includes: Data collection: Collect bus GPS trajectory data, bus stop latitude and longitude, and POI latitude and longitude; Grid encoding: encode the longitude and latitude into a string of length n, where n is the grid precision; Multi-scale grid generation: Generate a multi-scale grid system to form a set of geo-hash grids with different spatial resolutions; Trajectory mapping: Map the GPS data reported by the vehicle's driving trajectory to a geohash grid of corresponding accuracy in time series to form a trajectory point sequence; Spatiotemporal network graph construction: Construct a spatiotemporal network graph where nodes are all occurrences of geohash grids; determine spatiotemporal relationship edges and matching relationship edges.

5. The robust prediction method for bus arrival time based on sparse GPS data with missing arrival information according to claim 1 is characterized in that: The method for determining the attributes of the relationship edges in the network graph in step 3 is as follows: Driving time calculation: Calculate driving time based on the vehicle's historical GPS trajectory sequence; Statistical analysis: For grids of different scales, the mean and standard deviation of travel time are calculated; Spatial and temporal feature acquisition: Obtain historical weather by geographic hash grid division; Divide morning and evening peaks by timestamp; mark dates as weekdays, weekends, or statutory holidays; Attribute storage: A sliding window mechanism is used to calculate the mean and standard deviation of factors affecting weather, morning and evening peaks, and holidays, and store them as edge attributes.

6. The robust prediction method for bus arrival time based on sparse GPS data with missing arrival information according to claim 1 is characterized in that: The specific process of using the trained model to predict the bus arrival time in step 4 is as follows: First, based on the real-time GPS data obtained by the vehicle and the GPS data of the target bus stop to be predicted, node matching is performed in a pre-built multi-scale spatiotemporal network graph to accurately locate the corresponding node; then, the attribute information of these nodes themselves and the attribute data of the relationship edges between nodes are extracted as input to the model, and finally the predicted arrival time value of the target bus stop to be predicted is output.

Images