A dynamic matching method for carpooling of online car-hailing based on travel time prediction

Abstract

The application provides a ride-hailing carpool dynamic matching method based on travel time prediction. The method comprises the following steps: obtaining a data set of ride-hailing orders; constructing a travel time prediction model based on interpretable machine learning; training the travel time prediction model using the data set and the interpretable machine learning model SHAP method to obtain a trained travel time prediction model; constructing a shared network; predicting the travel time of each order by using the trained travel time prediction model; matching carpool orders according to the on-demand carpool dynamic matching algorithm; and dynamically dispatching the matched carpool orders in the direction of carbon emission reduction. The method uses the demand information and location data of passengers, matches the qualified passenger orders through reasonable algorithms and optimization strategies, reduces the empty running and repeated running of vehicles through carpooling, improves the service efficiency of ride-hailing and reduces carbon emissions.

Description

Technical Field

[0001] This invention relates to the field of urban road traffic operation management technology, and in particular to a dynamic matching method for ride-hailing based on travel time prediction. Background Technology

[0002] With the development of GPS, mobile communication, and internet technologies, ride-hailing services have become the preferred mode of transportation for many people due to their efficiency and convenience. Ride-hailing platforms can achieve efficient and accurate matching of supply and demand resources by integrating passenger and vehicle travel information in real time. Addressing the low utilization rate of taxi seats, they can leverage the internet's capabilities to organize shared rides among multiple passengers. Ride-sharing can optimize passenger pick-up and drop-off times and locations using intelligent algorithms, predicting travel time and thus making vehicle routes more rational. Compared to regular ride-hailing, ride-sharing improves vehicle utilization, reduces travel distance and carbon emissions, and is an effective means of quickly alleviating traffic congestion, energy waste, and environmental pollution.

[0003] While some breakthroughs and innovations have been achieved in ride-hailing matching methods, several problems remain. First, travel time is often difficult to predict accurately due to complex traffic and weather conditions, and the "black box" nature of conventional machine learning algorithms makes predictions uninterpretable. Second, unreasonable order matching strategies result in some ride-hailing orders having low route overlap and long detours. Therefore, developing an effective, low-carbon-oriented dynamic matching method for ride-hailing is a pressing issue. Summary of the Invention

[0004] The embodiments of the present invention provide a dynamic matching method for ride-hailing based on trip time prediction, so as to effectively improve the efficiency of ride-hailing services and reduce carbon emissions.

[0005] To achieve the above objectives, the present invention adopts the following technical solution.

[0006] A dynamic matching method for ride-sharing based on trip time prediction includes:

[0007] Obtain the dataset of ride-hailing orders;

[0008] A travel time prediction model based on interpretable machine learning is constructed. The travel time prediction model is trained using the SHAP method of the interpretable machine learning model with the dataset to obtain the trained travel time prediction model.

[0009] A shared network is built, and the travel time of each order is predicted by a trained travel time prediction model. The ride-sharing orders are matched according to the on-demand ride-sharing dynamic matching algorithm.

[0010] Driven by carbon emission reduction, dynamic order dispatching is implemented for ride-sharing matching.

[0011] Preferably, the dataset for obtaining ride-hailing orders includes:

[0012] Prepare and analyze ride-hailing order data, passenger pick-up and drop-off point of interest (POI) information, and corresponding weather information within a specific area. Clean the data according to the 3σ principle (sense of difference between different points of interest and different time), removing erroneous order information caused by location or time discrepancies. Add POI information and weather information near the pick-up and drop-off points to the order data. Simultaneously, match passenger pick-up and drop-off points to the nearest intersections based on the latitude and longitude information provided by the passengers.

[0013] 70% of the processed order data was used as the training dataset, and the remaining 30% of the order data was used as the test set.

[0014] Preferably, the construction of a travel time prediction model based on interpretable machine learning, which involves training the travel time prediction model using the SHAP method on the dataset to obtain a trained travel time prediction model, includes:

[0015] A travel time prediction model using the Stacking-XCL ensemble algorithm is constructed, employing XGBoost, CatBoost, and LGBM models as base learners and LinearRegression as a meta-learner. The training data for the Stacking-XCL ensemble algorithm consists of ride-hailing order data from the training dataset. The input data includes passenger pick-up and drop-off locations, trip distance, pick-up time, POI information for pick-up and drop-off locations, and weather information for the day. The output feature is the trip time for the order. The travel time prediction model is validated using the SHAP (Simplified Method Interpretive Machine Learning) method with K-fold cross-validation based on ride-hailing order data from the validation dataset. Each base learner in the Stacking-XCL ensemble algorithm is trained separately using both the training and validation datasets, yielding prediction results from multiple base learners. These prediction results are then fused using a meta-learner to obtain the final prediction result. When the error in the training dataset no longer decreases, the current model parameters are saved, resulting in the trained travel time prediction model.

[0016] Preferably, each base learner in the Stacking-XCL ensemble algorithm is trained using the training dataset and validation dataset respectively, resulting in prediction results from multiple different base learners. The prediction results from each base learner are then fused using a meta-learner to obtain the final prediction result, including:

[0017] When training the base learners in the Stacking-XCL ensemble algorithm, the K-fold cross-validation method is used to combine the prediction results of all base learners as training data for the meta-learner. The meta-learner uses the Linear Regression model and the Shapley method to learn and integrate the prediction results of each base learner.

[0018] In interpretable machine learning, the SHAP method provides an interpretation of travel time prediction models by assigning relative importance to input features that influence the model output. For a predicted value f with N features... x The Shapley value is calculated by adding various features to the travel time prediction model and comparing the model outputs with and without those features. For a specific feature n, the Shapley value Φ is calculated. n (f,x) is:

[0019]

[0020] Where N is the number of predicted features, and S is a subset of the features already added to the model. x This is the prediction result obtained using subset S. Φ n (f,x) is the Shapley value of feature n, Φ n (f,x) is the Shapley value of feature n. The importance of the feature is calculated by summing the magnitudes of the Shapley values ​​of all samples of the feature.

[0021] When using multiple base learners, the Stacking-XCL model's meta-learner trains a regression model, which then learns the weights of each base learner. The predictions from each base learner are then weighted to obtain the final prediction. During the training of the Stacking-XCL model, the weights of the base learners in the final prediction are adjusted based on their performance and importance.

[0022] Preferably, the construction of the shared network, which involves predicting the travel time of each order using a trained travel time prediction model and matching ride-sharing orders according to an on-demand ride-sharing dynamic matching algorithm, includes:

[0023] Suppose there exists a ride-hailing service platform that provides ride-sharing services to users. Passengers send their travel requests to the platform, and each request r... i∈R is defined as a tuple O i D represents the starting point of order i. i Indicates the endpoint of order i. Indicates the time when order i is requested to be sent out. Let O' represent the expected arrival time of order i, where R represents the demand collected by the system over a certain time interval. Based on latitude and longitude information, passenger pick-up and drop-off points are matched to the nearest intersection. Considering ride-sharing and matching of two orders, O''''''''''''''''''''''''''''''"""""' ... i and D′ i Let i and j represent the actual origin and destination after combining them. This represents the expected departure time for order i. This represents the expected departure time for order j;

[0024] For each order request, it is determined whether ride-sharing is possible based on the time window constraint δ and the passenger's allowed delay time Δ. Only when there are other orders within the time window and the departure and arrival times of the two passengers are within the allowable delay range are these two orders identified as ride-sharing orders. If order i and order j satisfy either the "first to board, first to alight" or "first to board, last to alight" ride-sharing method, order i and order j are added to the candidate set as potential ride-sharing orders.

[0025] Calculate the total delay, time saving, and mileage saving for each potential ride-sharing order in the candidate set. The estimated arrival time for trip i is... Calculate the actual arrival times of trips i and j based on equations 8 and 9. The actual arrival time of trip i is... It is determined by the departure time of trip i. O i With O j Travel time between O j With D i The actual arrival time of trip j is calculated from the travel time between the two trips. D needs to be added. i With D j Total delays during the journey This refers to the delay caused by the matching order of the drop-off points in the two trips, as described in Equation 10. (Saving time) This refers to the sum of the individual trip times of the two trips minus the total trip time after ride-sharing, calculated using Equation 11, to save mileage. The waiting time is calculated using Equation 12. This refers to the waiting time for the second passenger relative to the actual boarding time, calculated using Formula 13;

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032] In the formula This indicates the estimated departure time for trip i. Let i be the estimated arrival time of trip i. The actual arrival time of trip i is calculated using formula 8, T(O i O j T(O) represents the travel time between the boarding location of trip i and the boarding location of trip j. j D i T(D) represents the travel time between the pick-up point and the drop-off point of trip i. i D j ) represents the travel time between the drop-off point of trip i and the drop-off point of trip j, Γ(O i D i ) represents the distance between the travel point i and the pick-up and drop-off points, Γ(O i O j () represents the distance between the boarding locations of trip i and trip j;

[0033] In the order travel network, the weights of the edges are the total delay, time saved, and mileage saved for the calculated ride-sharing orders. With the objectives of minimizing the total delay and maximizing the total time and mileage saved after order merging, the maximum weight matching algorithm in graph theory is used to determine which orders can be merged, thus obtaining the ride-sharing order matching results.

[0034] Preferably, the dynamic dispatching of carpooling orders guided by carbon emission reduction includes:

[0035] Traffic emissions are calculated for aggregated orders using the COPERT III model, categorizing them into hot start emissions, cold start emissions, and evaporative emissions, with a hot start emission factor EF. k The thermal emission factor EF refers to the total amount of pollutants generated per kilometer of travel by a specific pollutant k in a single vehicle. k It has five parameters, namely a, b, c, d, and e, and its calculation method is Equation 14:

[0036]

[0037] CO2 and PM 2.5 The emission factor is directly proportional to FC fuel consumption, and is calculated using equations (15) and (16):

[0038]

[0039]

[0040] For each ride-hailing vehicle, without considering its return trip after completing its last ride, the amount of a specific pollutant k generated by it during all its rides N is represented as E1 (from the starting point to the first ride), E2 (each ride), and E3 (between rides).

[0041]

[0042] In Equation 17: For E1, consider the pollutant emissions k generated by a single ride-hailing vehicle i completing all its tasks N. We can use the intersection closest to the starting point of the first trip as the starting point. This indicates its heat emission factor in the process. For E2, the distance between the two points is denoted as E2. This represents its heat emission factor in the nth travel task. Let E3 be the trajectory distance between the origin and destination points of the nth trip; for E3, This represents the heat emission factor of the road segment between the m-th and m+1-th travel tasks, i.e., the road segment between the end point of the m-th travel task and the start point of the m+1-th travel task. Let be the distance between the m-th and m+1-th travel tasks;

[0043] The vehicle dispatching problem for ride-hailing services is described as dispatching a series of ride-hailing vehicles to serve users' travel needs within a certain period of time. Considering the emission of various pollutants calculated by Equation 17, the minimum path coverage problem in the order carpooling matching network is solved using the Hopcroft-Karp graph matching algorithm. Each path represents the orders continuously served by the same ride-hailing vehicle. Real-time dynamic vehicle dispatching and route planning are completed to obtain the optimal dispatching scheme.

[0044] As can be seen from the technical solutions provided by the embodiments of the present invention above, the present invention proposes a dynamic matching and carbon emission quantification method for ride-hailing carpooling based on trip time prediction, aiming to solve the problems of order matching and vehicle scheduling in the ride-hailing carpooling process. Simultaneously, by combining interpretable machine learning-based trip time prediction methods, the system utilizes passenger demand information and location data, and through reasonable algorithms and optimization strategies, matches passenger orders that meet the conditions. By reducing empty runs and redundant trips through carpooling trips, the system improves the efficiency of ride-hailing services and reduces carbon emissions.

[0045] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of the invention. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of a dynamic matching method for ride-hailing based on trip time prediction provided by an embodiment of the present invention;

[0048] Figure 2 A schematic diagram illustrating the training process of a Stacking-XCL model for travel time prediction provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram illustrating a two-order carpooling travel strategy provided by an embodiment of the present invention;

[0050] Figure 4 This invention provides a weekday order volume distribution map for ride-hailing services at various stations.

[0051] Figure 5 This invention provides a spatial distribution map of ride-hailing vehicle origin-destination (OD) volume at different times during a weekday.

[0052] Figure 6 This invention provides a distribution map of passenger travel time and travel distance for ride-hailing services on weekdays and weekends, as part of an embodiment of the invention.

[0053] Figure 7 This invention provides a distribution map of the carrying and vacant times and the carrying and vacant distances of ride-hailing vehicles.

[0054] Figure 8 This is a comparison chart of predicted and actual values ​​of a travel time prediction model provided in an embodiment of the present invention.

[0055] Figure 9 This is a global feature importance map of a travel time prediction model provided in an embodiment of the present invention;

[0056] Figure 10 This is a schematic diagram illustrating the local interpretability of a travel time prediction model provided in an embodiment of the present invention;

[0057] Figure 11 This is a schematic diagram of a 5-minute time granularity result for order carpooling matching provided by an embodiment of the present invention;

[0058] Figure 12 This invention provides an embodiment of the distribution of ride-hailing vehicle speeds and pollutant indicators generated over different time periods. Detailed Implementation

[0059] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0060] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0061] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0062] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0063] This invention provides a dynamic matching method for ride-hailing carpooling based on trip time prediction. This method analyzes travel behavior from massive online order data and combines it with interpretable machine learning-based trip time prediction to optimize ride-hailing carpooling, improving emission reduction benefits while ensuring economic efficiency. By constructing a shared network of carpooling and dispatching networks, an on-demand dynamic matching algorithm is proposed, and a graph matching algorithm is used to optimize dynamic dispatching of ride-hailing vehicles. This method mainly includes the following three modules:

[0064] Data analysis and trip time prediction module: By analyzing travel behavior in massive online order data, this module explores the characteristics and influencing factors of ride-hailing services and proposes a trip time prediction method based on interpretable machine learning to achieve dynamic prediction of ride-hailing trip times.

[0065] On-demand ride-sharing dynamic matching algorithm module: Considering dynamic demand response collaborative matching strategies, it constructs a shared network that combines carpooling network and dispatch network, and uses the trip time prediction results to dynamically match passengers in order to realize on-demand ride-sharing.

[0066] Carbon emission reduction optimization module: Guided by carbon emission reduction, an optimization model for dynamic dispatching of ride-hailing services is proposed. The optimal computational efficiency solution is obtained through graph matching algorithm to reduce vehicle carbon emissions and fleet size.

[0067] This invention provides a processing flow for a ride-hailing carpooling dynamic matching method based on trip time prediction, as follows: Figure 1 As shown, the processing steps include the following:

[0068] Step S1: Data preparation and statistical analysis.

[0069] This study prepared and analyzed Didi Chuxing ride-hailing order data, passenger pick-up and drop-off point (POI) information, and corresponding Beijing weather information for the area within Beijing's Fifth Ring Road in July 2018. First, the data was cleaned according to the 3σ principle for trip time and distance. Statistical analysis showed that ride-hailing trip time and distance roughly follow a normal distribution. According to the 3σ principle in statistics, outliers are defined as values ​​that deviate from the mean by more than three standard deviations. Under the assumption of a normal distribution, values ​​outside three standard deviations from the mean have a very low probability of occurrence and can be considered outliers. Therefore, cleaning the trip time and distance data using the 3σ principle can eliminate erroneous order information caused by location or time deviations. Subsequently, POI information near the pick-up and drop-off points, along with weather information such as temperature, weather conditions, and wind speed, were added to the order data. Simultaneously, based on the latitude and longitude information provided by passengers, the pick-up and drop-off points were matched to the nearest intersections to simplify the ride-sharing order modeling process and provide a basis for selecting ride-sharing orders. After constructing the data features, 70% of the order data is used as the training dataset, and the remaining 30% is divided into the test set.

[0070] Step S2: Construct a travel time prediction model based on interpretable machine learning. Use the SHAP (Shapley Additive Explanations) method to train the travel time prediction model to obtain the trained travel time prediction model.

[0071] Figure 2This diagram illustrates the training process of a Stacking-XCL model for travel time prediction, as provided in an embodiment of the present invention. The present invention employs XGBoost, CatBoost, and LGBM models as base learners and a LinearRegression model as a meta-learner to construct a travel time prediction model using the Stacking-XCL ensemble algorithm. This model is then used to predict the travel time of ride-hailing orders. The training data for the Stacking-XCL ensemble algorithm consists of ride-hailing order data from the training dataset processed in step S1. Input features include passenger pick-up and drop-off locations, travel distance, pick-up time, POI information for pick-up and drop-off locations, and weather information for the day. The output feature is the travel time of the order. Subsequently, the travel time prediction model was validated using the SHAP (Shapley Additive Explanations, an interpretable machine learning model) method through a K-fold cross-validation process based on the ride-hailing order data in the validation dataset. K is 5, meaning that the base learners of the Stacking-XCL model are trained using 4 of the 5 equally divided data from the training dataset and validated using 1 of the data. This process is repeated 5 times to obtain 5 prediction values ​​for different validation datasets. Finally, the prediction results of each base learner on the validation dataset are fused through a meta-learner. The trained travel time prediction model is obtained when the error in the training dataset no longer decreases.

[0072] The Stacking-XCL ensemble algorithm excels at handling large-scale data and complex features, providing accurate predictions. K-fold cross-validation is used during base learner training to avoid overfitting. The predictions from all base learners are combined as training data for the meta-learner. The meta-learner uses a Linear Regression model, learning and integrating the predictions from each base learner to obtain a more accurate final prediction. To enhance the interpretability of the travel time prediction model, the SHAP method, based on the Shapley value concept from game theory, is used to explain the impact of various features of a single sample on the prediction result in complex algorithms.

[0073] The Shapley value is a commonly used method in interpretable machine learning to evaluate the contribution of features to model predictions. Based on the concept of Shapley values ​​in cooperative game theory, it aims to assign a relative weight to each feature to explain the model's predictions, calculated as shown in Equation 1. In interpretable machine learning, the Shapley value determines the degree of influence of each feature on the model output, i.e., the importance of the feature. It quantifies the contribution of each feature to the model's predictions by calculating the interaction between each feature and other features. The SHAP method provides an explanation for travel time prediction models by assigning relative importance to input features that affect the model's output. For a predicted value f with N features... x The Shapley value is calculated by adding one feature after another to the travel time prediction model and comparing the outputs of models with and without those features. Therefore, for a specific feature n, the Shapley value Φ... n (f,x) is:

[0074]

[0075] Where N is the number of predicted features, and S is a subset of the features already added to the model. x This is the prediction result obtained using subset S. Φ n (f,x) is the Shapley value of feature n, Φ n (f,x) is the Shapley value of feature n. Since the prediction may depend on the features added first, an average feature ranking is performed on the contributions of all (N!) possible features. To compute this most efficiently, this invention uses the SHAP package, which calculates feature importance by summing the magnitudes of the Shapley values ​​for all samples of a feature. When using multiple base learners, the meta-learner of the Stacking-XCL model is trained as a regression model to learn the weights of each base learner, and the predictions of each base learner are weighted to obtain the final prediction. The Stacking-XCL model employs a model fusion strategy to automatically learn weights, adjusting the weights (contribution levels) of the base learners in the final prediction based on their performance and importance, to obtain more accurate and reliable ensemble predictions.

[0076] Unlike traditional feature importance assessment methods, the Shapley score can not only assess the global importance of features, but also provide specific interpretability for each case locally. It offers more accurate and transparent interpretability, which is of great help to decision-makers in understanding the model.

[0077] Step S3: Construct a shared network, predict the travel time of each order using a trained travel time prediction model, and match ride-sharing orders according to the on-demand ride-sharing dynamic matching algorithm.

[0078] Suppose there exists a ride-hailing service platform that provides carpooling services to users. Passengers send their travel requests to the platform, and each request r... i ∈R can be defined as a tuple O i D represents the starting point of order i. i Indicates the endpoint of order i. Indicates the time when order i is requested to be sent out. Let R represent the expected arrival time of order i, where R represents the demand collected by the system at a certain time interval, such as 5 minutes. Based on the submitted travel information, ride-sharing conditions are determined, and a matching solution is derived through optimization. To perform efficient matching calculations, this invention matches passenger pick-up and drop-off points to the nearest intersection based on latitude and longitude information. This invention considers ride-sharing and matching of two orders, O′ i and D′ i These represent the actual origin and destination after combining orders i and j, respectively. This represents the expected departure time for order i; This indicates the expected departure time for order j.

[0079] For each order request, the feasibility of ride-sharing is determined based on the time window constraint (δ) and the passenger's allowed delay time (Δ). Ride-sharing is only considered possible if other orders exist within the time window and both passengers' departure and arrival times are within the allowable delay range. Some parameters are defined as follows:

[0080]

[0081]

[0082] Simultaneously considering both "first-come, first-served (FAFL)" and "first-come, first-served (FALL)" ride-sharing methods, an order meeting either of these methods is considered a potential ride-sharing order and added to the candidate set. For example, in the attached... Figure 3 In (a), the two orders satisfy the ride-sharing constraint, and the vehicle will first arrive at the starting point O of order i. i Pick up the first passenger, then at the starting point O of order j. j Pick up another passenger. The vehicle first arrives at destination D′ of order i. i After completing the delivery of the first passenger, it then proceeds to the destination D′ of order j. j The other passenger is then delivered. In this case, order i is the first order in the ride-sharing order list, so its expected pick-up time equals its actual pick-up time. Order j, however, is affected by the ride-sharing arrangement, and its expected pick-up time is later than its actual pick-up time. The condition for this situation to be satisfied is:

[0083]

[0084]

[0085]

[0086]

[0087] Equation 2 indicates that the expected pick-up time for both orders must not exceed the time window constraint; Equation 3 indicates that the time when the vehicle arrives at the pick-up location of the second passenger after picking up the first passenger can be earlier than the actual pick-up time of the second passenger, i.e., the vehicle waits for the passenger, but cannot be later than the latest expected pick-up time of the second passenger, i.e., the expected pick-up time. The passenger's allowed delay Δ is added; Equations 4 and 5 respectively indicate that the vehicle's arrival time at the first and second passenger's drop-off point must not be later than the latest arrival time.

[0088] In the appendix Figure 3 In (b), the vehicle will be at the starting point O of order i. i Pick up the passenger, and then at the starting point O of order j j Pick up another passenger. The vehicle first arrives at destination D′ of order j. j After completing the delivery of the first passenger, it then proceeds to the destination D′ of order j. i The second passenger has been delivered. At this point, conditions 4 and 5 are changed to 6 and 7:

[0089]

[0090]

[0091] Calculate the total delay, time savings, and mileage savings for each potential ride-sharing order pair in the candidate set. The estimated arrival time for trip i is... Based on equations 8 and 9, the actual arrival times of trips i and j can be calculated, and the actual arrival time of trip i is... It is determined by the departure time of trip i. O i With O j Travel time between O j With D i The actual arrival time of trip j is calculated from the travel time between the two trips. D needs to be added. i With D j Travel time between destinations. Total delays. This refers to the delay caused by the matching order of the drop-off points in the two trips, as described in Equation 10. (Saving time) This refers to the sum of the individual trip times of the two trips minus the total trip time after the ride-sharing arrangement, as shown in Equation 11. Similarly, mileage savings... Waiting time can be calculated according to Equation 12. This refers to the waiting time for the second passenger relative to the actual boarding time, calculated using Formula 13.

[0092] In the formula This indicates the estimated departure time for trip i. Let i be the estimated arrival time of trip i. The actual arrival time of trip i can be calculated using formula 8. T(O) i O j T(O) represents the travel time between the boarding location of trip i and the boarding location of trip j. j D i T(D) represents the travel time between the pick-up point and the drop-off point of trip i. i D j Γ(O) represents the travel time between the drop-off point of trip i and the drop-off point of trip i. i D i ) represents the distance between the travel point i and the pick-up and drop-off points, Γ(O i O j ) represents the distance between the boarding locations of trip i and trip j. For detailed parameter descriptions, please refer to the variable description in step S3.

[0093] Based on the carpooling matching objectives such as the maximum number of matched carpooling orders, the minimum total delay of carpooling orders, and the maximum time and mileage savings, the maximum weight matching algorithm in graph theory is used to obtain the order carpooling matching results. First, an order travel network is constructed, representing ride-hailing orders as nodes and edges as carpooling matching relationships between orders. Considering both "first-come, first-served" (FAFL) and "first-served, last-served" (FALL) carpooling methods, orders satisfying formulas 2-7 are identified as potential carpooling orders and added to the candidate set. The total delay, time savings, and mileage savings for each pair of potential carpooling orders in the candidate set are calculated using formulas 8-13. Then, the ride-hailing order carpooling matching can be solved using the maximum weight matching algorithm in graph theory. That is, in the order travel network, the weights of the edges are the calculated total delay, time savings, and mileage savings of the carpooling orders. With the objectives of minimizing the total delay after order merging and maximizing the total time and mileage savings, the maximum weight matching algorithm in graph theory is used to determine the order carpooling matching results. Finally, based on the results of the maximum weight matching algorithm, it is determined which orders can be merged, and a ride-sharing matching scheme is derived.

[0094]

[0095]

[0096]

[0097]

[0098]

[0099]

[0100] Step S4: Dynamically dispatch carpooling orders based on carbon emission reduction.

[0101] The Hopcroft-Karp algorithm, used for graph matching, is employed to obtain the optimal dispatch plan based on passenger demand and vehicle availability. The ride-hailing vehicle dispatch problem can be described as dispatching a series of ride-hailing vehicles to serve user travel needs within a given time period. Research has shown that for directed acyclic networks, finding the minimum path cover problem is equivalent to the maximum matching problem on a bipartite graph, which can be solved using the Hopcroft-Karp algorithm. In graph theory, the Hopcroft-Karp algorithm is an efficient algorithm for solving the maximum matching problem. It expands the size of partial matchings by repeatedly finding augmenting paths, i.e., by switching edges on the path, adding or removing them from the partial matching. The Hopcroft-Karp algorithm expands the matching using the maximum set of shortest augmenting paths, where these shortest augmenting paths have the same length and a maximum length. This algorithm has low time complexity and good scalability and practicality.

[0102] Traffic emissions were then calculated using the COPERT III model. The COPERT III model is a widely used model for traffic emissions estimation, capable of accurately estimating and predicting transportation emissions based on factors such as vehicle technical characteristics, speed, driving mode, and road characteristics. The COPERT III model divides traffic emissions into three parts: hot start emissions, cold start emissions, and evaporative emissions. Under normal driving conditions, hot start emissions are the primary source of engine emissions; therefore, the latter two, which are of less importance, are ignored in the estimation process. The thermal emission factor (EF) is used. k () refers to the total amount of pollutants (grams / km) generated by a specific pollutant k per kilometer traveled by a single vehicle. It is calculated using Equation 14, is related to driving speed (km / h), and has five parameters: a, b, c, d, and e.

[0103]

[0104] The pollution emission parameters in the COPERT III model are shown in Table 1:

[0105] Table 1 Pollution emission parameters in the COPERT III model

[0106]

[0107] As for CO2 and PM 2.5 Other pollutants, such as fuel consumption (FC), have emission factors that are directly proportional to their emission factors, for example, as shown in the following formula:

[0108]

[0109]

[0110] For each ride-hailing vehicle, without considering its return trip after completing its last ride, the amount of a specific pollutant k generated by it during all its rides N can be expressed as the amount generated by the ride-hailing vehicle from its origin to its first ride, E1; the amount generated during each ride, E2; and the amount generated between two consecutive rides, E3.

[0111]

[0112] In Equation 17: Let E1 be the pollutant emissions (k) generated by a single ride-hailing vehicle i completing all its tasks N. For E1, we consider the intersection closest to the starting point of the first trip as the starting point. This indicates its heat emission factor in the process. For E2, the distance between the two points is denoted as E2. This represents its heat emission factor in the nth travel task. Let E3 be the trajectory distance between the origin and destination points of the nth trip; for E3, This represents the heat emission factor of the route between the m-th and m+1-th travel tasks (i.e., the route between the end point of the m-th travel task and the start point of the m+1-th travel task). Let be the distance between the m-th and m+1-th travel tasks. In our study, k represents different types of emission pollutants.

[0113] During the dispatching process, vehicle routes and trips are adjusted and optimized in real time to minimize carbon emissions and fleet size. In practice, the ride-hailing platform dynamically processes order data, for example, every 5 minutes. First, it uses the trip time prediction model proposed in step S2 to predict the trip time of the order. Then, it uses the ride-sharing dynamic matching method proposed in step S3 to match the order demands collected by the platform, constructing a shared network. Finally, considering minimizing the various pollutant emissions calculated in Equation 17, the Hopcroft-Karp graph matching algorithm is used to solve the minimum path coverage problem in the order ride-sharing matching network. Each path represents orders continuously served by the same ride-hailing vehicle, completing real-time dynamic vehicle dispatching and route planning.

[0114] Example 1

[0115] Reference Figure 1 The processing steps of the ride-hailing carpooling dynamic matching method based on trip time prediction provided in this embodiment of the invention include:

[0116] Step S1: Data Preparation and Analysis. This invention requires the following data: Didi Chuxing order data within Beijing's Fifth Ring Road in July 2018, intersection data and surrounding POI information within Beijing's Fifth Ring Road, and Beijing weather information in July 2018. Data cleaning was performed on trip time and distance according to the 3σ principle, ultimately using 14,259,826 data points for the experiment. The data characteristics of Beijing ride-hailing services include: driver ID, passenger ID, pick-up and drop-off time, pick-up and drop-off location, passenger travel distance (km), and travel time (s). This invention matches pick-up and drop-off locations to the nearest intersection and adds surrounding POI information. This invention also considers the impact of weather factors on ride-hailing trips and order matching, using weather conditions, maximum temperature, minimum temperature, and wind speed to describe the weather characteristics of the day. (Appendix) Figure 4 This describes the time distribution of ride-hailing orders at various stations on weekdays, with attached... Figure 5 The spatial distribution of ride-hailing origin-destination (OD) data at different times during weekdays is described, with appendix. Figure 6 The kernel density estimation distribution of ride-hailing vehicle occupancy time and distance over the week is described, with appendix. Figure 7 It describes the distribution of ride-hailing vehicle occupancy and vacancy times, as well as the distances between occupancy and vacancy.

[0117] Step S2: Analyze travel behavior in online order data based on the statistically processed ride-hailing order data, and train a travel time prediction model based on interpretable machine learning.

[0118] This paper applies the Stacking-XCL strategy, using XGBoost, CatBoost, and LGBM models as base learners, to predict the travel time of ride-hailing services in Beijing during weekdays and weekends, and peak and off-peak hours, respectively. The mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R²) are then used as the statistical measures. 2 Three evaluation indicators are used to evaluate the predictive performance of the model. The calculation methods for the three evaluation indicators are shown in formulas 18, 19, and 20:

[0119]

[0120]

[0121]

[0122] Tables 2 and 3 show the comparison of prediction accuracy of the Stacking-XCL model with other baseline models during weekdays and weekends, and peak and off-peak hours, respectively. Figure 8 The prediction accuracy of the Stacking-XCL model and other baseline models on weekdays and weekends was visualized. Observation and analysis of the results lead to the conclusion that the Stacking-XCL model, which uses XGBoost, CatBoost, and LGBM models as base learners, can accurately predict ride-hailing travel times, significantly outperforming other baseline models, and the results are reliable. In the dataset, the average travel time for all trips is 1005 seconds, and the standard deviation is 603. The mean squared error (RMSE) of the travel times predicted by the Stacking-XCL model ranges from a minimum of 208.3 to a maximum of 235; the mean absolute error (MAE) ranges from a minimum of 146 to a maximum of 169; and the coefficient of determination (R²) is [not specified in the original text]. 2 The accuracy can reach 0.8779. The results show that the Stacking-XCL model has high accuracy in predicting travel times across various scenarios.

[0123] Table 2. Prediction accuracy of the travel time prediction model on weekdays and weekends.

[0124]

[0125] Table 3. Prediction accuracy of the travel time prediction model during peak and off-peak hours.

[0126]

[0127]

[0128] The SHAP method was used in the experiment to interpret the machine learning model. Figure 9This visually illustrates the impact and effectiveness of each feature on the prediction results from a global perspective. The number of scatter points for each feature represents the number of samples, the scatter range represents the range of variation in the feature's SHAP value, and the color indicates the magnitude of the feature value. All variables are arranged in descending order of impact, with top-ranked variables contributing more to the prediction model than bottom-ranked variables. In terms of impact, travel distance has the greatest influence on prediction, while wind level has the least. In terms of effectiveness, the greater the travel distance, the greater the increase in travel time compared to the average; conversely, the smaller the travel distance, the greater the decrease in travel time compared to the average. (Appendix) Figure 10 This study demonstrates the local interpretability of three typical samples. Calculating the SHAP value of each feature for a single sample more clearly shows how the model uses data features to predict a sample, providing a more intuitive understanding. The comparison reveals that, using the same average as a baseline, different features of different samples have different impacts, pushing the prediction results towards different values. On the one hand, the same feature can produce different effects and degrees of influence; on the other hand, the features that influence the predictions of different samples also differ. In practical application, it is easy to see that the different impacts of different features in the model are consistent with real-world situations.

[0129] Step S3: In this embodiment of the invention, based on the order merging model proposed in Step S3, the time window limit δ is set to 120 seconds, and the maximum allowable delay Δ is 120 seconds. Based on the two ride-sharing methods "First to Board, First to Alight (FAFL)" and "First to Board, Last to Alight (FALL)," orders that satisfy one of these methods are considered potential ride-sharing orders and added to the candidate set.

[0130] To evaluate the effectiveness of order merging every 5 minutes, the following experiment was conducted: For each potential ride-sharing order pair, its total delay, time saved, mileage saved, and waiting time were calculated. Based on the ride-sharing matching results every 5 minutes, an hourly average was calculated to reflect the matching effectiveness of the order-sharing algorithm within that hour.

[0131] Table 4 shows the hourly average results of ride-sharing matching during peak hours on weekdays and weekends. The results indicate that ride-hailing order volume is slightly lower during weekday peak hours than on weekends. Specifically, the highest order volume on weekends occurs at 9:00 and 19:00, later than 8:00 and 17:00 on weekdays, suggesting that ride-hailing services are primarily used for early commuting during weekday morning and evening peak hours. Merging potential orders through ride-sharing can effectively reduce the number of orders. During weekday peak hours, the order volume can be reduced by up to 34.18%; during weekend peak hours, the order volume can be reduced by up to 34.10%. Furthermore, the ride-sharing matching algorithm achieves significant time and mileage savings through lower delays and passenger waiting times. During the 5:00 PM time slot on weekdays, the delay is only 6.65 seconds per 5 minutes, and the second passenger's additional waiting time increases by only 27.44 seconds, while saving 841.56 seconds of time and 4223.36 meters of distance. During the 6:00 PM time slot on weekends, the delay is only 6.04 seconds per 5 minutes, and the second passenger's additional waiting time increases by only 29.70 seconds, while saving 716.82 seconds of time and 3503.14 meters of distance.

[0132] Table 4 Results of Optimized Order Matching During Peak Hours

[0133]

[0134] Figure 11 This is a heatmap showing ride-sharing results every 5 minutes on weekdays. The vertical axis represents a 24-hour day, and the horizontal axis represents 12 five-minute intervals per hour. Each sub-plot displays different ride-sharing metrics. (Attached) Figure 11 (a) Displays the actual number of ride-hailing orders, with appendix Figure 11 (b) Displaying the number of orders after ride-sharing matching clearly shows that ride-sharing effectively reduced the number of orders. (Appendix) Figure 11 (c) Displays delay information after ride-sharing matching. Clearly, delays caused by order matching are low in most periods, having minimal impact on order punctuality. However, during the early morning hours, due to lower order volume, matching with distant orders may result in significant delays. (Appendix) Figure 11 (d) and appendix Figure 11 (e) represents the time and mileage saved after ride-sharing matching. The degree of time and mileage savings is roughly the same, but during peak hours, due to road congestion, mileage savings are greater than time savings; during off-peak hours, due to the more dispersed distribution of orders, time savings are greater than mileage savings. (Appendix) Figure 11 (f) shows the extra waiting time for the second passenger after ride-sharing. It can be seen that the passenger waiting time is relatively even, mostly around 30 seconds, and the longest waiting time does not exceed 1 minute. Compared with the time saved, the waiting time is within an acceptable range.

[0135] Step S4: Based on the original orders and ride-sharing orders, the Hopcroft-Karp algorithm is used for dynamic order dispatching, allowing drivers to continuously serve different orders. Considering the impact of vehicle speed on carbon emissions, the COPERT III model is used to quantify the carbon emissions before and after order merging. (Appendix) Figure 12 This describes the speed distribution of actual ride-hailing vehicles at different times and the pollution generated at different speeds. (See appendix.) Figure 12 In (a), ride-hailing vehicles travel at relatively high speeds between 6:00 and 8:00 and between 10:00 and 12:00, with a maximum speed of approximately 78 km / h. During the morning rush hour from 8:00 to 10:00, vehicle speeds are slower, with an average speed of only 19.72 km / h. (Appendix) Figure 12 (b) shows the impact of real-time driving speed on fuel consumption and emission coefficients. It can be seen that vehicle speed has a significant impact on carbon emissions, and variations in each speed range result in significant differences in carbon emissions. A driving speed of approximately 60 km / h is the speed at which vehicle fuel consumption is minimized and pollutant reduction is most efficient, while driving speeds below 20 km / h tend to result in higher pollution emissions.

[0136] Table 5 shows the results of vehicle dispatch optimization during peak weekday periods. Statistics are compiled in one-hour timeframes. At 7:00 AM, 8:00 AM, and 9:00 AM, the order demand for ride-sharing services was 2606, 1879, and 1903 respectively. Using a ride-sharing order merging strategy, these orders were reduced to 1915, 1453, and 1423 respectively. Without the order merging strategy, 2606 orders needed to be served between 7:00 AM and 8:00 AM, requiring 1006 ride-hailing vehicles to meet these passengers' travel needs, resulting in a total vehicle travel distance of 11461.21 km. Based on the actual driving speeds of different vehicles, the COPERT III model was used to calculate vehicle pollutant emissions. The total fuel consumption of all vehicles in one hour was 1129.11 kg, with emissions of 6068.94 g CO, 329.56 g HC, 995.28 g NOx, 3590.57 kg CO2, and PM2.5. 2.5 The amount of fuel consumed was 33.87g. When the order consolidation strategy was adopted, the number of consolidated orders was 1915, requiring only 684 vehicles to serve these orders, reducing the total ride-hailing trip distance by 25.17%. Correspondingly, the vehicle's pollutant emissions also decreased, with CO, HC, and NOx emissions reduced to 4437.94g, 234.33g, and 742.87g, respectively. The total fuel consumption of these vehicles decreased to 805.69kg, and CO2 and PM2.5 emissions also decreased. 2.5Emissions decreased by 28.64%. During the evening rush hour, passenger travel is more dispersed, and some large numbers of travel demands continue into the night. Between 7 PM and 8 PM, there were 1618 ride-hailing orders within Beijing's Third Ring Road, requiring 559 vehicles to serve these orders, with a total vehicle distance of 7778.48 km. After order consolidation, the number of orders decreased to 1221, requiring 406 vehicles to serve these orders, with a total vehicle distance of 6179.84 km, representing reductions of 27.37% and 20.55% in the number of vehicles and total distance, respectively. Regarding pollutant emissions, the reduction in ride-hailing demand led to a decrease in various pollutant indicators. After adopting the order consolidation strategy, vehicle NOx and CO2 emissions were 525.55g and 1658.69kg, respectively, compared to 665.37g and 2167.23g without the strategy, representing reductions of 21.01% and 23.47%.

[0137] Table 5 Results of vehicle allocation optimization during peak hours on weekdays

[0138]

[0139] Therefore, by comprehensively considering travel time prediction, ride-sharing order matching, and carbon emission quantification, we have achieved the goal of reducing the number of orders and pollutant emissions by optimizing algorithms and real-time data, and obtained low-carbon ride-sharing dynamic matching and dispatch results.

[0140] In summary, the method of this invention can improve the efficiency of ride-hailing services: by screening ride-sharing orders and optimizing vehicle scheduling, multiple orders can be combined into a ride-sharing trip, minimizing empty runs and duplicate trips, improving vehicle utilization, and thus improving the overall efficiency of ride-hailing services.

[0141] The method described in this invention can reduce carbon emissions: by establishing a shared network and rationally scheduling vehicles, carpooling trips can reduce vehicle mileage and pollutant emissions. This helps alleviate traffic congestion and reduce environmental pollution, contributing to sustainable development.

[0142] The method described in this invention can improve user experience: considering factors such as comfort and safety, ride-sharing order screening can ensure suitable matching between passengers, providing a more comfortable and safer travel experience. Furthermore, optimized matching and scheduling of ride-sharing orders also helps reduce passenger waiting time and improve travel efficiency.

[0143] The method described in this invention can match potential ride-sharing orders in real time and efficiently, optimize vehicle dispatching routes, and has significant advantages in improving the operational efficiency of ride-hailing services, increasing the utilization rate of urban roads, alleviating traffic congestion, and reducing traffic carbon emissions, thus promoting the sustainable development of urban transportation.

[0144] Flexible travel options: For passengers, the ride-sharing order filtering provides a flexible travel choice. When suitable, passengers can choose ride-sharing to enjoy a more economical and convenient travel service. Even when no suitable ride-sharing orders are available, passengers can still post non-ride-sharing trips to ensure their travel needs are met.

[0145] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0146] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0147] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the embodiments of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0148] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A dynamic matching method for ride-hailing carpooling based on trip time prediction, characterized in that, include: Obtain the dataset of ride-hailing orders; A travel time prediction model based on interpretable machine learning is constructed. The travel time prediction model is trained using the SHAP method of the interpretable machine learning model with the dataset to obtain the trained travel time prediction model. A shared network is built, and the travel time of each order is predicted by a trained travel time prediction model. The ride-sharing orders are matched according to the on-demand ride-sharing dynamic matching algorithm. Driven by carbon emission reduction, dynamic order dispatching is implemented for ride-sharing matching orders; The construction of a travel time prediction model based on interpretable machine learning, which involves training the model using the SHAP method on the dataset to obtain a trained travel time prediction model, includes: A travel time prediction model using the Stacking-XCL ensemble algorithm is constructed, employing XGBoost, CatBoost, and LGBM models as base learners and a Linear Regression model as a meta-learner. The training data for the Stacking-XCL ensemble algorithm consists of ride-hailing order data from the training dataset. The input data for the Stacking-XCL ensemble algorithm includes passenger pick-up and drop-off locations, travel distance, pick-up time, POI information for pick-up and drop-off locations, and the weather information for the day. The output feature is the travel time for that order. Based on ride-hailing order data from the validation dataset, the interpretable machine learning model SHAP is used to predict travel time. The cross-validation process validates the travel time prediction model. Each base learner in the Stacking-XCL ensemble algorithm is trained using the training dataset and the validation dataset respectively, resulting in prediction results from multiple different base learners. The prediction results from each base learner are then fused through a meta-learner to obtain the final prediction result. When the error in the training dataset no longer decreases, the current model parameters are saved to obtain the trained travel time prediction model. In interpretable machine learning, the SHAP method provides an explanation for travel time prediction models by assigning relative importance to input features that influence the model output. Predicted values ​​of each feature The Shapley value is calculated by adding various features to the travel time prediction model and comparing the outputs of models with and without those features. Shapley value for: in, To predict the number of features, It is a subset of the features that have already been added to the model. Is using this subset The predicted results obtained It is a feature The Shapley value is used to calculate the importance of a feature by summing the magnitudes of the Shapley values ​​for all samples of that feature.

2. The method according to claim 1, characterized in that, The dataset for obtaining ride-hailing orders includes: Prepare and analyze the order data of ride-hailing vehicles in a specific area, the points of interest information of passenger pick-up and drop-off locations, and the corresponding weather information. Clean the data according to the 3σ principle for trip time and trip distance, and remove erroneous order information caused by location or time deviation. Add points of interest information and weather information near the pick-up and drop-off points to the order data. At the same time, match the passenger's pick-up and drop-off points to the nearest intersection based on the latitude and longitude information provided by the passenger. 70% of the processed order data was used as the training dataset, and the remaining 30% of the order data was used as the test set.

3. The method according to claim 2, characterized in that, Each base learner in the Stacking-XCL ensemble algorithm is trained using the training and validation datasets respectively, resulting in predictions from multiple different base learners. The predictions from these base learners are then fused using a meta-learner to obtain the final prediction result, including: When training the base learners in the Stacking-XCL ensemble algorithm, the following is adopted: The cross-validation method combines the predictions of all base learners as training data for the meta-learner. The meta-learner uses the Linear Regression model and the Shapley method to learn and integrate the predictions of each base learner. When using multiple base learners, the meta-learner of the Stacking-XCL model trains a regression model, which learns the weights of each base learner. The predictions of each base learner are weighted to obtain the final prediction result. During the training of the Stacking-XCL model, the weights of the base learners in the final prediction result are adjusted according to their performance and importance.

4. The method according to claim 3, characterized in that, The aforementioned construction of a shared network involves predicting the travel time of each order using a trained travel time prediction model, and matching ride-sharing orders according to an on-demand ride-sharing dynamic matching algorithm, including: Suppose there exists a ride-hailing service platform that provides carpooling services to users. Passengers send their travel requests to the platform, and each request... Defined as a tuple , Indicates order The starting point Indicates order The end point Indicates order The time when the request is issued, Indicates order The expected arrival time, of which This represents the demand collected by the system over a certain time interval. Based on latitude and longitude information, it matches passenger pick-up and drop-off points to the nearest intersection, considering ride-sharing and matching of two orders. These represent the actual origin and destination after the ride-sharing order; For each order requirement, based on the time window constraint and passenger-allowed delay time Δ To determine whether a ride-sharing arrangement is possible, two orders are only considered ride-sharing orders if there are other orders within the time window and the departure and arrival times of both passengers are within the allowable delay range; order judgment... and orders When the order meets either the "first-come, first-served" or "first-served, last-disembark" carpooling method, the order will be processed. and orders Added to the candidate set as a potential ride-sharing order; Calculate the total delay, time saved, and mileage saved for each potential ride-sharing order in the candidate set, and the trip details. The estimated arrival time is Calculate the travel distance according to equations 8 and 9. With itinerary Actual arrival time, itinerary actual arrival time It is a trip Departure time , Travel time between The travel time between the two points is calculated. actual arrival time Additional additions are needed Total delays during the journey This refers to the delay caused by the matching order of the drop-off points in the two trips in Equation 10, thus saving time. This refers to the sum of the individual trip times of the two trips minus the total trip time after ride-sharing, calculated using Equation 11, to save mileage. The waiting time is calculated using Equation 12. This refers to the waiting time for the second passenger relative to the actual boarding time, calculated using Formula 13; In the formula Indicates itinerary The estimated departure time, For the itinerary Estimated arrival time For the itinerary The actual arrival time is calculated using Formula 8. Indicates itinerary Pick-up location and itinerary The travel time between the pick-up points, Indicates itinerary Pick-up location and itinerary The travel time between the drop-off points Indicates itinerary Drop-off point and itinerary The travel time between the drop-off points Indicates itinerary The distance between the pick-up and drop-off points. Indicates itinerary Pick-up location and itinerary The distance between pick-up points; In the order travel network, the weights of the edges are the total delay, time saved, and mileage saved for the calculated ride-sharing orders. With the objectives of minimizing the total delay and maximizing the total time and mileage saved after order merging, the maximum weight matching algorithm in graph theory is used to determine which orders can be merged, thus obtaining the ride-sharing order matching results.

5. The method according to claim 4, characterized in that, The aforementioned dynamic dispatching of ride-sharing orders, guided by carbon emission reduction, includes: Traffic emissions for ride-sharing orders are calculated using the COPERT III model, categorizing them into hot-start emissions, cold-start emissions, and evaporative emissions. The hot-start emission factor... Refers to specific pollutants The total amount of pollutants generated per kilometer of travel by a single vehicle, thermal emission factor It has five parameters, namely Its calculation method is Equation 14: （14） The emission factor is directly proportional to FC fuel consumption, and is calculated using equations (15) and (16): For each ride-hailing vehicle, without considering its return trip after completing its last ride, the completion of all rides... Specific pollutants produced The quantity is represented as the number of ride-hailing trips generated from the origin to the first trip. The amount of tasks generated per trip The amount generated between two consecutive travel missions : In Equation 17: For a single ride-hailing vehicle Complete all its tasks The pollutants produced k Emissions, for Consider using the intersection closest to the starting point of the first travel task as the starting point. This indicates its heat emission factor in the process. The distance between two points; for Indicates that it is in the first Thermal emission factors in a travel mission For the first The trajectory distance between the origin and destination points of each travel task; for Indicates that it is in the first The route between each travel task, i.e. the first The destination of the second trip and the first Heat emission factors of the road segment between the starting points of the next trip. In the first The distance between each travel task; The vehicle dispatching problem for ride-hailing services is described as dispatching a series of ride-hailing vehicles to serve users' travel needs within a certain period of time. Considering the emission of various pollutants calculated by Equation 17, the minimum path coverage problem in the order carpooling matching network is solved using the Hopcroft-Karp graph matching algorithm. Each path represents the orders continuously served by the same ride-hailing vehicle. Real-time dynamic vehicle dispatching and route planning are completed to obtain the optimal dispatching scheme.

Images