Online car-hailing order late reminding method and system

Abstract

The invention relates to the technical field of online car-hailing order processing, in particular to a late reminding method and system for an online car-hailing order. According to the method, the order information of the online car-hailing driver is recorded, and the online car-hailing order information set is constructed; the online car-hailing order information set comprises online car-hailing time information and order feature information; screening the order feature information to obtain order features; according to the order features, adopting a plurality of late time prediction single models to predict late time, and obtaining a plurality of late time single prediction values; inputting the plurality of late time single predicted values into a late time variable weight combination prediction model to obtain a late time predicted value; and judging whether the driver is late or not according to the output late time predicted value, and when the judgment result is late, transmitting order late information to a passenger and reminding the passenger. Through the late reminding method of the online car-hailing order, the late time of the online car-hailing driver can be predicted more accurately.

Description

Technical Field

[0001] This invention relates to the field of ride-hailing order processing technology, specifically to a method and system for providing late arrival reminders for ride-hailing orders. Background Technology

[0002] Ride-hailing services have become an indispensable part of urban life, providing passengers with convenient transportation options. However, with the rapid growth and progress of this sector, some challenges have emerged, one of which is the issue of late orders.

[0003] In traditional ride-hailing services, passengers often find it difficult to accurately know the driver's arrival time, which can lead to inconvenience. Existing ride-hailing services typically only provide rough arrival time estimates. For example, Chinese patent application CN202110689735.X provides a method, device, terminal, and storage medium for vehicle lateness warning. After a ride order is generated, it obtains the order recipient's historical lateness rate, current location, and the order initiator's departure location and departure time. If the order recipient's historical lateness rate falls within a preset high lateness rate range, it plans a travel route from the order recipient's current location to the order initiator's departure location and calculates the corresponding estimated travel time. Based on the order recipient's estimated travel time and the order initiator's departure time, it determines whether the order recipient can arrive at the order initiator's departure location on time. A warning is issued to order recipients who cannot arrive on time. However, this method does not consider real-time road conditions, which can easily lead to errors in judging whether an order will be late. Furthermore, Chinese patent application number CN202211476661.2 provides a method and device for late arrival reminders for ride-hailing orders. This method obtains the driver's ID, location, current time, the ID of the driver's next pending order, the scheduled pick-up time of the driver's next pending order ID, and the scheduled pick-up point location of the driver's next pending order ID. If pending orders exist, it calculates the shortest pick-up route distance and travel time for the next pending order. Based on historical order data of the driver's location area, it calculates the tolerance time for the driver's location area. Based on the tolerance time for the driver's location area, the shortest pick-up route distance, and travel time for the driver's next pending order, it generates late arrival reminder rules. However, this method does not consider real-time road conditions and other factors, and its considerations are not comprehensive enough, which can easily lead to inaccurate predictions of ride-hailing driver lateness.

[0004] To address the problem of inaccurate predictions of ride-hailing drivers' lateness, a method and system for providing lateness reminders for ride-hailing orders are proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for reminding passengers of late arrivals for ride-hailing orders. By collecting information from ride-hailing orders, multiple individual prediction models for late arrival times are used to predict the late arrival situation and late arrival time. Then, a combined prediction model for late arrival times is used to integrate the results of the individual predictions to obtain the final predicted value of the late arrival time. Based on the predicted value, a reminder message is sent to the passenger.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for issuing late arrival reminders for ride-hailing orders includes:

[0008] Record the historical order information of ride-hailing drivers and construct a ride-hailing order information set.

[0009] Furthermore, the ride-hailing time information includes the planned arrival time at the pick-up location and the actual arrival time at the pick-up location; the lateness time is obtained by subtracting the planned arrival time from the actual arrival time at the pick-up location; when the lateness time is positive, it indicates that the order is late; otherwise, it indicates that the order is not late; the order feature information includes location information, traffic condition information, driver information, and weather information; the location information includes the distance of the first straight line from the driver's location when accepting the order to the pick-up location and the length of the first actual route, the distance of the second straight line from the pick-up location to the destination location and the length of the second actual route; the traffic condition information includes the number of traffic lights on the first actual route, whether the first actual route contains traffic jams, and the cumulative distance of the traffic jams; the driver information includes the driver's gender, age, driving experience, and total number of orders accepted; the weather information includes whether it is extreme weather.

[0010] The order feature information is filtered to obtain the order features.

[0011] Furthermore, the filtering of the order feature information includes:

[0012] The Pearson correlation coefficient and Spearman correlation coefficient were used to filter the late arrival characteristic information. The specific steps included:

[0013] Calculate the Pearson correlation coefficient and Spearman correlation coefficient for each of the lateness features and lateness times;

[0014] Remove order features whose Pearson correlation coefficient and Spearman correlation coefficient are both between [-0.2, 0.2]; output order features.

[0015] Furthermore, the Pearson correlation coefficient takes values ​​of [-1, 1], where a Pearson correlation coefficient of 1 indicates a perfect positive correlation, a Pearson correlation coefficient of -1 indicates a perfect negative correlation, and a Pearson correlation coefficient of 0 indicates no correlation; the Spearman correlation coefficient takes values ​​of [-1, 1], where a Spearman correlation coefficient of 1 indicates a perfect positive correlation, a Spearman correlation coefficient of -1 indicates a perfect negative correlation, and a Spearman correlation coefficient of 0 indicates no correlation.

[0016] Furthermore, the Pearson correlation coefficient is used to verify the linear relationship between the two factors, and the Spearman correlation coefficient is used to verify the nonlinear relationship between the two factors.

[0017] The order features are input into multiple late time prediction models to obtain multiple late time prediction values.

[0018] Furthermore, the late arrival time prediction single model includes a late arrival LSTM model:

[0019] The parameters in the late LSTM model are optimized using an improved dung beetle optimization algorithm.

[0020] The parameters include the number of hidden units, the learning rate, and the batch size;

[0021] The specific steps include:

[0022] Initialize a dung beetle swarm, where each dung beetle position in the swarm represents a set of parameter values;

[0023] The prediction accuracy between the predicted value of the late time output by the late LSTM model and the actual value of the late time is calculated based on the location of each dung beetle, and the negative number of the prediction accuracy is used as the fitness value of each dung beetle.

[0024] The dung beetle colony was divided into five subgroups in equal proportions: rolling, dancing, breeding, foraging, and stealing.

[0025] The positions of all dung beetles are adaptively updated based on their fitness values.

[0026] The position update formula for the rolling dung beetle is:

[0027] x i (t+1)=x i (t)+β*h*x i (t-1)+b*|x i (t)-X ω |;

[0028] Where t represents the current iteration number; x i(t) represents the position information of the i-th dung beetle in the t-th iteration; h∈(0,0.2] represents the deflection coefficient; b∈(0,1); β is the natural coefficient, assigned a value of -1 or 1; X ω Indicates the worst-case position globally;

[0029] The position update formula for the dancing dung beetle is:

[0030] x i (t+1)=x i (t)+tanθ|x i (t)-x i (t-1)|;

[0031] Where θ∈[0,π] represents the deflection angle, if Then the dung beetle's location is updated;

[0032] The breeding dung beetles reproduce and lay eggs in a specific area, which is:

[0033]

[0034] Among them, X * Lb represents the current local optimum. * and Ub * Let Lb and Ub represent the lower and upper bounds of the spawning zone, respectively; T represents the maximum number of iterations; Lb and Ub represent the lower and upper bounds of the optimization problem, respectively; the formula for updating the position of the spawn ball is:

[0035] B i (t+1)=X * +b1×(B i (t)-Lb * )+b2×(B i (t)-Ub * );

[0036] Among them, B i (t) represents the position information of the i-th egg in the t-th iteration, and b1 and b2 represent two independent 1×3 random vectors;

[0037] The location of the foraging dung beetle has been updated to:

[0038] x i (t+1)=x i (t)+e -t / T ×(x i (t)-Lb b )+e -t / T ×(x i (t)-Ub b );

[0039] Among them, Lbb and Ub b Let Lb represent the upper and lower bounds of the optimal foraging region, respectively. b and Ub b The expressions are as follows:

[0040]

[0041] Among them, X b This represents the globally optimal foraging location; the position update expression for the dung beetle is:

[0042] x i (t+1)=X b +S×b×(|x i (t)-X * |+|x i (t)-X b |);

[0043] Where b represents a 1×3 random vector, and S represents a constant value.

[0044] Determine whether each dung beetle has exceeded the boundary;

[0045] Update the current optimal solution and its fitness value;

[0046] Repeat the above steps until the set number of iterations is reached, and output the global optimal solution of the parameters in the late LSTM model.

[0047] Furthermore, the steps for the late arrival time prediction using the late arrival LSTM model include:

[0048] The order information set was divided into an order training set and an order test set in a 7:3 ratio.

[0049] Initialize the late LSTM model;

[0050] The order features from the late arrival training set are input into the late arrival LSTM model. An improved dung beetle optimization algorithm is used to optimize the parameters in the late arrival LSTM model, and the late arrival time prediction value is output.

[0051] Calculate the lateness loss function between the predicted value of the lateness time and the actual value of the lateness time;

[0052] If the late loss function converges, output the trained late LSTM model; otherwise, continue training.

[0053] The expression for the late arrival loss function is:

[0054]

[0055] Where m1 represents the number of late orders in the training set; C i This represents the actual late time of the i-th order in the late training set; Let η represent the predicted late time of the i-th order in the late training set; η > 0 is a smoothing term.

[0056] Input multiple individual predicted values ​​of late arrival time into the late arrival time variable weight combination prediction model to obtain the predicted value of late arrival time.

[0057] Furthermore, the late-arriving variable-weight combination prediction model is constructed based on the induced ordered weighted multiple average operator; the expression of the induced ordered weighted multiple average operator is:

[0058]

[0059] in, Represented as the predicted late time based on the induced ordered weighted multiple average operator; α1, α2, ..., α n Let η1, η2, ..., η be the individual predicted values ​​of late arrival time from n late arrival time prediction models; n ω represents the prediction accuracy of n individual models for predicting lateness; k Let α represent the weighting coefficient assigned to the k-th lateness prediction single model; η-index(k) It is represented as the late time prediction value corresponding to the kth prediction accuracy after the prediction accuracy is arranged in ascending order;

[0060] The expression for the late-arriving variable weight combination prediction model is:

[0061]

[0062] Where C represents the actual lateness time.

[0063] Based on the predicted lateness time, it is determined whether the driver will be late. If the determination result is that the driver will be late, the passenger is sent a late order information and a reminder is given.

[0064] Furthermore, the order lateness information includes the estimated lateness time and the probability of lateness; the probability of lateness is calculated based on the driver's historical late orders, and the expression is:

[0065]

[0066] Where P is the probability of being late, and D c D represents the number of historical late orders for the driver. z The total number of historical orders for the driver; the output of the prediction result based on the lateness characteristics of the current order also includes:

[0067] When the predicted result is late, the number of orders dispatched to the driver is adjusted according to the expected lateness and the probability of lateness, as specifically expressed by:

[0068]

[0069] Where f∈(0,1) represents the adjustment coefficient, and PDL is the number of orders dispatched by the driver before the adjustment. new This represents the number of orders dispatched to drivers after the adjustment, T. l This represents the set threshold.

[0070] A late arrival reminder system for ride-hailing orders, characterized in that it includes:

[0071] Order Record Unit: Used to record the historical order information of ride-hailing drivers and build a ride-hailing order information set;

[0072] Filtering unit: used to filter the order feature information to obtain order features;

[0073] Prediction unit: Used to predict the arrival time of ride-hailing drivers;

[0074] Decision unit: Used to determine whether the driver is late based on the predicted arrival time;

[0075] Output unit: Used to output late order information based on the discrimination result of the discrimination unit.

[0076] Furthermore, the prediction unit includes obtaining a predicted late time value based on the order characteristics using a late time prediction model. The prediction steps include:

[0077] The order features are input into multiple late time prediction models to obtain multiple late time prediction values.

[0078] Based on multiple individual prediction values ​​of lateness time, a lateness time prediction model is used to obtain the predicted value of lateness time.

[0079] The late-arriving variable weight combination prediction model is constructed based on the induced ordered weighted multiple average operator;

[0080] The expression for the induced ordered weighted multiple average operator is:

[0081]

[0082] in, Represented as the predicted late time based on the induced ordered weighted multiple average operator; α1, α2, ..., α n Let η1, η2, ..., η be the individual predicted values ​​of late arrival time from n late arrival time prediction models; nω represents the prediction accuracy of n individual models for predicting lateness; k Let α represent the weighting coefficient assigned to the k-th lateness prediction single model; η-index(k) It is represented as the late time prediction value of the single model corresponding to the kth prediction accuracy after the prediction accuracy is arranged in ascending order;

[0083] The expression for the late-arriving variable weight combination prediction model is:

[0084]

[0085] Where C represents the actual lateness time.

[0086] Furthermore, the late arrival time prediction single model includes:

[0087] Use the late LSTM model as a single model for predicting lateness time;

[0088] The parameters in the late LSTM model are optimized using an improved dung beetle optimization algorithm.

[0089] The parameters include the number of hidden units, the learning rate, and the batch size;

[0090] The specific steps include:

[0091] Initialize a dung beetle swarm, where each dung beetle position in the swarm represents a set of parameter values;

[0092] The prediction accuracy between the predicted late time output by the late LSTM model and the actual late time is calculated based on the location of each dung beetle, and the inverse of the prediction accuracy is used as the fitness value of each dung beetle.

[0093] The dung beetle colony was divided into five subgroups in equal proportions: rolling, dancing, breeding, foraging, and stealing.

[0094] The positions of all dung beetles are adaptively updated based on their fitness values.

[0095] The breeding dung beetles reproduce and lay eggs in a specific area, which is:

[0096]

[0097] Among them, X * Lb represents the current local optimum. * and Ub * Let Lb and Ub represent the lower and upper bounds of the spawning zone, respectively; T represents the maximum number of iterations; Lb and Ub represent the lower and upper bounds of the optimization problem, respectively; the formula for updating the position of the spawn ball is:

[0098] B i(t+1)=X * +b1×(B i (t)-Lb * )+b2×(B i (t)-Ub * );

[0099] Among them, B i (t) represents the position information of the i-th egg in the t-th iteration, and b1 and b2 represent two independent 1×3 random vectors;

[0100] The location of the foraging dung beetle has been updated to:

[0101] x i (t+1)=x i (t)+e -t / T ×(x i (t)-Lb b )+e -t / T ×(x i (t)-Ub b );

[0102] Among them, Lb b and Ub b Let Lb represent the upper and lower bounds of the optimal foraging region, respectively. b and Ub b The expressions are as follows:

[0103]

[0104] Among them, X b This is indicated as the optimal foraging location globally;

[0105] Determine whether each dung beetle has exceeded the boundary;

[0106] Update the current optimal solution and its fitness value;

[0107] Repeat the above steps until the set number of iterations is reached, and output the global optimal solution of the parameters in the late LSTM model.

[0108] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0109] 1. When predicting the lateness time of ride-hailing orders, this invention combines location information, traffic condition information, driver information, and weather information, and uses multiple lateness time prediction models to make individual predictions of the lateness time. Then, based on the induced ordered weighted multiple average operator and the Softmax function, a lateness time variable weight combination prediction model is constructed. The multiple lateness time prediction values ​​are input into the lateness time variable weight combination prediction model. Compared with the existing lateness time prediction models, the variable weight combination prediction model can better integrate the prediction advantages of each lateness time prediction model and can more accurately predict the lateness time.

[0110] 2. This invention employs an improved dung beetle optimization algorithm to optimize the parameters of the single model for late arrival time prediction. Based on the number of iterations and the prediction accuracy between the predicted and actual values ​​of late arrival time, the algorithm adaptively updates the positions of breeding and foraging dung beetles within the population, selecting the optimal dung beetle position as the optimal solution for the parameters. Compared to the dung beetle optimization algorithm, the improved algorithm finds the optimal solution faster and more accurately, improving the accuracy of the single model for late arrival time prediction, and thus improving the overall accuracy of late arrival time prediction.

[0111] 3. This invention, based on the LSTM model, employs an improved dung beetle optimization algorithm to optimize the number of hidden units, learning rate, and batch size of the LSTM, and uses the late arrival loss function as a criterion for verifying the model's training quality, constructing a late arrival LSTM model. The late arrival loss function can capture the asymmetric difference between the predicted and actual late arrival times, and also adapts to the case where the late arrival time is negative. Using the late arrival LSTM model as the single model for late arrival time prediction improves the prediction accuracy of the LSTM, thereby enhancing the overall accuracy of late arrival time prediction. Attached Figure Description

[0112] Figure 1 This is a schematic diagram of a method for issuing late reminders for ride-hailing orders, provided in an embodiment of the present invention.

[0113] Figure 2 This is a schematic diagram of the Pearson correlation coefficient heatmap provided in an embodiment of the present invention;

[0114] Figure 3 A schematic diagram of the Spearman correlation coefficient heatmap provided in an embodiment of the present invention;

[0115] Figure 4 This is a comparison chart of the performance of the late LSTM model provided in the embodiments of the present invention;

[0116] Figure 5 A comparison chart of the performance of the late BP neural network model provided in the embodiments of the present invention;

[0117] Figure 6 This is a comparison chart of the performance of the late random forest model provided in the embodiments of the present invention;

[0118] Figure 7 Comparison chart of the effects of the late XGBoost model provided in the embodiments of the present invention;

[0119] Figure 8 A schematic diagram illustrating the prediction of late arrival time provided in an embodiment of the present invention;

[0120] Figure 9 This is a comparison chart of the performance of the late-arriving variable-weight combination prediction model provided in the embodiments of the present invention;

[0121] Figure 10 This is a schematic diagram of a late arrival reminder system for ride-hailing orders provided in an embodiment of the present invention. Detailed Implementation

[0122] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0123] Please see Figures 1 to 10 This invention provides a method and system for late arrival reminders for ride-hailing orders, the technical solution of which is as follows:

[0124] As one embodiment of the present invention, refer to Figure 1 A method for reminding passengers of late arrivals for ride-hailing orders includes: S10. Recording the historical order information of ride-hailing drivers to construct a ride-hailing order information set; S20. Filtering the order feature information to obtain order features; S30. Inputting the order features into multiple late arrival time prediction models to obtain multiple late arrival time prediction values; S40. Inputting the multiple late arrival time prediction values ​​into a late arrival time variable weight combination prediction model to obtain late arrival time prediction values; S5. Determining whether the driver will be late based on the late arrival time prediction values, and when the determination result is late, sending late arrival information to the passenger and reminding them.

[0125] S10. Record the historical order information of ride-hailing drivers and construct a ride-hailing order information set.

[0126] Furthermore, in this embodiment, a total of 2,000 historical order information from different ride-hailing drivers were collected. The ride-hailing time information includes the planned arrival time and the actual arrival time at the pick-up location. The late time is obtained by subtracting the planned arrival time from the actual arrival time at the pick-up location. When the late time is positive, it indicates that the order is late; otherwise, it indicates that the order is not late. The order feature information includes location information, traffic condition information, driver information, and weather information. The location information includes the distance of the first straight line from the driver's location when accepting the order to the pick-up location and the length of the first actual route, the distance of the second straight line from the pick-up location to the destination location, and the length of the second actual route. The traffic condition information includes the number of traffic lights on the first actual route, whether the first actual route contains traffic jams, and the cumulative distance of the traffic jams. The driver information includes the driver's gender, age, driving experience, and total number of orders accepted. The weather information includes whether it is extreme weather.

[0127] S20. Filter the order feature information to obtain order features.

[0128] Furthermore, the filtering of the order feature information includes:

[0129] The Pearson correlation coefficient and Spearman correlation coefficient were used to filter the late arrival characteristic information. The specific steps included:

[0130] Calculate the Pearson correlation coefficient and Spearman correlation coefficient for each of the lateness features and lateness times;

[0131] Order feature information where both the Pearson correlation coefficient and the Spearman correlation coefficient are between [-0.2, 0.2] is removed;

[0132] Output order characteristics.

[0133] Furthermore, the Pearson correlation coefficient takes values ​​of [-1, 1], where a Pearson correlation coefficient of 1 indicates a perfect positive correlation, a Pearson correlation coefficient of -1 indicates a perfect negative correlation, and a Pearson correlation coefficient of 0 indicates no correlation; the Spearman correlation coefficient takes values ​​of [-1, 1], where a Spearman correlation coefficient of 1 indicates a perfect positive correlation, a Spearman correlation coefficient of -1 indicates a perfect negative correlation, and a Spearman correlation coefficient of 0 indicates no correlation.

[0134] Furthermore, the Pearson correlation coefficient is used to verify the linear relationship between the two factors, and the Spearman correlation coefficient is used to verify the nonlinear relationship between the two factors.

[0135] Furthermore, based on the calculated results of the Pearson correlation coefficient and the Spearman correlation coefficient, the following plot is drawn: Figure 2and Figure 3 The correlation coefficient graph shown is from Figure 2 and Figure 3 It can be seen that the correlation coefficients between the distance of the second straight line and the length of the second actual route and the late time and other order feature information are all between [-0.2, 0.2], so these two order feature information are removed.

[0136] S30. Input the order features into multiple late time prediction models to obtain multiple late time prediction values.

[0137] Furthermore, the late arrival time prediction single model includes:

[0138] Use the late LSTM model as a single model for predicting lateness time;

[0139] The late LSTM model uses an improved dung beetle optimization algorithm to optimize the parameters in the late LSTM.

[0140] The parameters include the number of hidden units, the learning rate, and the batch size;

[0141] The specific steps include:

[0142] Initialize a dung beetle swarm, where each dung beetle position in the swarm represents a set of parameter values;

[0143] The prediction accuracy between the predicted late time output by the late LSTM model and the actual late time is calculated based on the location of each dung beetle, and the negative value of the prediction accuracy is used as the fitness value of each dung beetle.

[0144] The dung beetle colony was divided into five subgroups in equal proportions: rolling, dancing, breeding, foraging, and stealing.

[0145] The positions of all dung beetles are adaptively updated based on their fitness values.

[0146] The position update formula for the rolling dung beetle is:

[0147] x i (t+1)=x i (t)+β*h*x i (t-1)+b*|x i (t)-X ω |;

[0148] Where t represents the current iteration number; x i (t) represents the position information of the i-th dung beetle in the t-th iteration; h∈(0,0.2] represents the deflection coefficient; b∈(0,1); β is the natural coefficient, assigned a value of -1 or 1; X ω Indicates the worst-case position globally;

[0149] The position update formula for the dancing dung beetle is:

[0150] x i (t+1)=x i (t)+tanθ|x i (t)-x i (t-1)|;

[0151] Where θ∈[0,π] represents the deflection angle, if Then the dung beetle's location is updated;

[0152] The breeding dung beetles reproduce and lay eggs in a specific area, which is:

[0153]

[0154] Among them, X * Lb represents the current local optimum. * and Ub * Let Lb and Ub represent the lower and upper bounds of the spawning zone, respectively; T represents the maximum number of iterations; Lb and Ub represent the lower and upper bounds of the optimization problem, respectively; the formula for updating the position of the spawn ball is:

[0155] B i (t+1)=X * +b1×(B i (t)-Lb * )+b2×(B i (t)-Ub * );

[0156] Among them, B i (t) represents the position information of the i-th egg in the t-th iteration, and b1 and b2 represent two independent 1×3 random vectors;

[0157] The location of the foraging dung beetle has been updated to:

[0158] x i (t+1)=x i (t)+e -t / T ×(x i (t)-Lb b )+e -t / T ×(x i (t)-Ub b );

[0159] Among them, Lb b and Ub b Let Lb represent the upper and lower bounds of the optimal foraging region, respectively. b and Ub b The expressions are as follows:

[0160]

[0161] Among them, X b This represents the globally optimal foraging location; the position update expression for the dung beetle is:

[0162] x i (t+1)=X b +S×b×(|x i (t)-X * |+|x i (t)-X b |);

[0163] Where b represents a 1×3 random vector, and S represents a constant value.

[0164] Determine whether each dung beetle has exceeded the boundary;

[0165] Update the current optimal solution and its fitness value;

[0166] Repeat the above steps until the set number of iterations is reached, and output the global optimal solution of the parameters in the late LSTM model.

[0167] Furthermore, the steps for the late arrival time prediction using the late arrival LSTM model include:

[0168] The order information set was divided into an order training set and an order test set in a 7:3 ratio.

[0169] Initialize the late LSTM model;

[0170] The order features in the late training set are input into the late LSTM model. An improved dung beetle optimization algorithm is used to optimize the parameters in the late LSTM model, and the late time prediction value of the late LSTM model is output.

[0171] Calculate the lateness loss function between the predicted lateness time and the actual lateness time;

[0172] If the late loss function converges, output the trained late LSTM model; otherwise, continue training.

[0173] The expression for the late arrival loss function is:

[0174]

[0175] Where m1 represents the number of late orders in the training set; C i This represents the actual late time of the i-th order in the late training set; Let η represent the predicted late time of the i-th order in the late training set; η = 2 -0.6 This is a smoothing term.

[0176] Furthermore, this embodiment employs an improved dung beetle optimization algorithm to optimize the learning rate, number of hidden layer nodes, number of hidden layers, and batch size of the BP neural network model, and uses the late arrival loss function as the prediction loss function for the BP neural network model to construct a late arrival BP neural network model; it also employs an improved dung beetle optimization algorithm to optimize the number of trees, maximum depth, minimum number of leaf node samples, and maximum number of features of the random forest model, and uses the late arrival loss function as the prediction loss function for the random forest model to construct a late arrival random forest model; and it employs an improved dung beetle optimization algorithm to optimize the learning rate, number of trees, maximum depth, minimum sum of subsample weights, column sampling ratio, and subsample ratio of the XGBoost model, and uses the late arrival loss function as the prediction loss function for the random forest model to construct a late arrival XGBoost model. The late arrival BP neural network model, late arrival random forest model, and late arrival XGBoost model are used as single models for late arrival time prediction to perform single-item predictions of late arrival time, obtaining multiple single-item prediction values ​​for late arrival time.

[0177] In addition to using late LSTM, late BP neural network, late random forest, and late XGBoost models for single-item prediction, this embodiment also replaces the dung beetle optimization algorithm with an improved one to optimize the parameters of the LSTM, BP neural network, random forest, and XGBoost models, constructing a first LSTM model, a first BP neural network model, a first random forest model, and a first XGBoost model for single-item prediction of lateness time. Furthermore, the LSTM, BP neural network, random forest, and XGBoost models are used for single-item prediction of lateness time. Each model is tested 20 times repeatedly, and the average error value of each single model for predicting lateness time in the test set is calculated for each experiment. The results are as follows. Figures 4-7 As shown. By Figures 4-7 It can be seen that the late LSTM model, late BP neural network model, late random forest model, and late XGBoost model constructed in this embodiment can effectively improve the prediction performance of the single model for late time prediction.

[0178] S40. Input multiple individual prediction values ​​of late arrival time into the late arrival time variable weight combination prediction model to obtain the prediction value of late arrival time.

[0179] Furthermore, the late-arriving variable-weight combination prediction model is constructed based on the induced ordered weighted multiple average operator; the expression of the induced ordered weighted multiple average operator is:

[0180]

[0181] in, Represented as the predicted late time based on the induced ordered weighted multiple average operator; α1, α2, ..., α n Let η1, η2, ..., η be the individual predicted values ​​of late arrival time from n late arrival time prediction models; n ω represents the prediction accuracy of n single models for predicting late arrival times. During real-time prediction, the average prediction accuracy of each single model for predicting late arrival times for a ride-hailing driver in the last three predictions is used as the prediction accuracy of that single model during the actual prediction. k Let α represent the weighting coefficient assigned to the k-th lateness prediction single model; η-index(k) It is represented as the late time prediction value corresponding to the kth prediction accuracy after the prediction accuracy is arranged in ascending order;

[0182] The expression for the late-arriving variable weight combination prediction model is:

[0183]

[0184] Where C represents the actual lateness time; S() represents the value of the Softmax function;

[0185] Furthermore, refer to Figure 8 The prediction results of four single models for predicting late arrival times were input into a variable-weighted combination prediction model for late arrival times. The predicted late arrival times were calculated using the formula of the variable-weighted combination prediction model. The prediction results of 20 repeated trials of the four single models for predicting late arrival times were then input into the variable-weighted combination prediction model for late arrival times. The prediction error of the variable-weighted combination prediction model was calculated based on the results of the variable-weighted combination prediction model for late arrival times and compared with the prediction errors of the four single models for predicting late arrival times. Specific results are shown below. Figure 9 As shown. According to Figure 9 The results show that the prediction performance of the late arrival variable weight combination prediction model provided in this embodiment is significantly higher than that of the four late arrival time prediction single models.

[0186] S50. Based on the predicted lateness time, determine whether the driver will be late. If the determination result is that the driver will be late, send the late order information to the passenger and remind them.

[0187] Furthermore, the order lateness information includes the estimated lateness time and the probability of lateness; the probability of lateness is calculated based on the driver's historical late orders, and the expression is:

[0188]

[0189] Where P is the probability of being late, and D c D represents the number of historical late orders for the driver. zThe total number of the driver's historical orders; the step of determining whether the driver will be late based on the predicted lateness time also includes:

[0190] When the predicted result is late, the number of orders dispatched to the driver is adjusted according to the expected lateness and the probability of lateness, as specifically expressed by:

[0191]

[0192] Where f∈(0,1) represents the adjustment coefficient, and PDL is the number of orders dispatched by the driver before the adjustment. new This represents the number of orders dispatched to drivers after the adjustment, T. l This represents the set threshold.

[0193] As one embodiment of the present invention, refer to Figure 10 A late arrival reminder system for ride-hailing orders, characterized by comprising: S210. an order recording unit for recording historical order information of ride-hailing drivers and constructing a ride-hailing order information set; S220. a filtering unit for filtering the order feature information to obtain order features; S230. a prediction unit for predicting the arrival time of ride-hailing drivers; S240. a discrimination unit for determining whether the driver is late based on the predicted arrival time; and S250. an output unit for outputting late arrival information based on the discrimination result of the discrimination unit.

[0194] Furthermore, the prediction unit includes obtaining a predicted late time value based on the order characteristics using a late time prediction model. The prediction steps include:

[0195] The order features are input into multiple late time prediction models to obtain multiple late time prediction values.

[0196] Based on multiple individual prediction values ​​of lateness time, a lateness time prediction model is used to obtain the predicted value of lateness time.

[0197] The late-arriving variable weight combination prediction model is constructed based on the induced ordered weighted multiple average operator;

[0198] The expression for the induced ordered weighted multiple average operator is:

[0199]

[0200] in, Represented as the predicted late time based on the induced ordered weighted multiple average operator; α1, α2, ..., α n Let η1, η2, ..., η be the individual predicted values ​​of late arrival time from n late arrival time prediction models; nω represents the prediction accuracy of n individual models for predicting lateness; k Let α represent the weighting coefficient assigned to the k-th lateness prediction single model; η-index(k) It is represented as the late time prediction value of the single model corresponding to the kth prediction accuracy after the prediction accuracy is arranged in ascending order;

[0201] The expression for the late-arriving variable weight combination prediction model is:

[0202]

[0203] Where C represents the actual lateness time; S() represents the value of the Softmax function.

[0204] Furthermore, the late arrival time prediction single model includes a late arrival LSTM model:

[0205] The parameters in the late LSTM model are optimized using an improved dung beetle optimization algorithm.

[0206] The parameters include the number of hidden units, the learning rate, and the batch size;

[0207] The specific steps include:

[0208] Initialize a dung beetle swarm, where each dung beetle position in the swarm represents a set of parameter values;

[0209] The prediction accuracy between the predicted value of the late time output by the late LSTM model and the actual value of the late time is calculated based on the location of each dung beetle, and the negative number of the prediction accuracy is used as the fitness value of each dung beetle.

[0210] The dung beetle colony was divided into five subgroups in equal proportions: rolling, dancing, breeding, foraging, and stealing.

[0211] The positions of all dung beetles are adaptively updated based on their fitness values.

[0212] The breeding dung beetles reproduce and lay eggs in a specific area, which is:

[0213]

[0214] Among them, X * Lb represents the current local optimum. * and Ub * Let Lb and Ub represent the lower and upper bounds of the spawning zone, respectively; T represents the maximum number of iterations; Lb and Ub represent the lower and upper bounds of the optimization problem, respectively; the formula for updating the position of the spawn ball is:

[0215] B i (t+1)=X *+b1×(B i (t)-Lb * )+b2×(B i (t)-Ub * );

[0216] Among them, B i (t) represents the position information of the i-th egg in the t-th iteration, and b1 and b2 represent two independent 1×3 random vectors;

[0217] The location of the foraging dung beetle has been updated to:

[0218] x i (t+1)=x i (t)+e -t / T ×(x i (t)-Lb b )+e -t / T ×(x i (t)-Ub b );

[0219] Among them, Lb b and Ub b Let Lb represent the upper and lower bounds of the optimal foraging region, respectively. b and Ub b The expressions are as follows:

[0220]

[0221] Among them, X b This is indicated as the optimal foraging location globally;

[0222] Determine whether each dung beetle has exceeded the boundary;

[0223] Update the current optimal solution and its fitness value;

[0224] Repeat the above steps until the set number of iterations is reached, and output the global optimal solution of the parameters in the late LSTM model.

[0225] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0226] One or more embodiments of this specification are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the scope of protection of this disclosure.

Claims

1. A method for providing late arrival reminders for ride-hailing orders, characterized in that, include: Record ride-hailing drivers' order information and construct a ride-hailing order information set; The ride-hailing order information set includes ride-hailing time information and order characteristic information; Filter the order feature information to obtain order features; The order features are input into multiple late time prediction models to obtain multiple late time prediction values. Input multiple individual prediction values ​​of lateness time into the lateness variable weight combination prediction model to obtain the prediction value of lateness time; The late-arriving variable weight combination prediction model is constructed based on the induced ordered weighted multiple average operator; The expression for the induced ordered weighted multiple average operator is: in, Represented as the predicted late time based on the induced ordered weighted multiple average operator; α1, α2, ..., α n Let η1, η2, ..., η be the individual predicted values ​​of late arrival time from n late arrival time prediction models; n ω represents the prediction accuracy of n individual models for predicting lateness; k Let α represent the weighting coefficient assigned to the k-th lateness prediction single model; η-index(k) It is represented as the late time prediction value corresponding to the kth prediction accuracy after the prediction accuracy is arranged in ascending order; The expression for the late-arriving variable weight combination prediction model is: Where C represents the actual lateness time; Based on the predicted lateness time, it is determined whether the driver will be late; if the determination result is that the driver will be late, the passenger is sent a late order information and a reminder is given.

2. The method for late arrival reminders for ride-hailing orders according to claim 1, characterized in that, The ride-hailing order information set includes: The ride-hailing time information includes the planned arrival time at the pick-up location and the actual arrival time at the pick-up location. The lateness time is obtained by subtracting the planned arrival time from the actual arrival time at the boarding location. If the late time is positive, it means the order is late; otherwise, it means the order is not late. The order feature information includes location information, traffic condition information, driver information, and weather information; The location information includes the distance of the first straight line from the driver's location when accepting the order to the pick-up location and the length of the first actual route, and the distance of the second straight line from the pick-up location to the destination location and the length of the second actual route. The traffic condition information includes the number of traffic lights on the first actual route, the number of congested sections on the first actual route, and the cumulative distance of the congested sections; The driver information includes the driver's gender, age, driving experience, and total number of orders received; The weather information includes whether it is extreme weather.

3. The method for late arrival reminders for ride-hailing orders according to claim 1, characterized in that, The order feature information to be filtered includes: The Pearson correlation coefficient and Spearman correlation coefficient were used to filter the late arrival characteristic information. The specific steps included: Calculate the Pearson correlation coefficient and Spearman correlation coefficient for each of the lateness features and lateness times; Remove order features whose Pearson correlation coefficient and Spearman correlation coefficient are both between [-0.2, 0.2]; output order features.

4. The method for late arrival reminders for ride-hailing orders according to claim 1, characterized in that, The single model for predicting late arrival time includes a late arrival LSTM model: The parameters in the late LSTM model are optimized using an improved dung beetle optimization algorithm. The parameters include the number of hidden units, the learning rate, and the batch size; The specific steps include: Initialize a dung beetle swarm, where each dung beetle position in the swarm represents a set of parameter values; The prediction accuracy between the predicted late time output by the late LSTM model and the actual late time is calculated based on the location of each dung beetle, and the inverse of the prediction accuracy is used as the fitness value of each dung beetle. The dung beetle colony was divided into five subgroups in equal proportions: rolling, dancing, breeding, foraging, and stealing. Based on the fitness value, select the dung beetle in the local best position and the dung beetle in the global best position, and update the position of all dung beetles; The breeding dung beetles reproduce and lay eggs in a specific area, which is: Among them, X * This represents the current local optimum; Lb * and Ub * Let Lb and Ub represent the lower and upper bounds of the spawning zone, respectively; T represents the maximum number of iterations; Lb and Ub represent the lower and upper bounds of the optimization problem, respectively; the formula for updating the position of the spawn ball is: B i (t+1)=X * +b1×(B i (t)-Lb * )+b2×(B i (t)-Ub * ); Among them, B i (t) represents the position information of the i-th egg in the t-th iteration, and b1 and b2 represent two independent 1×3 random vectors; The location of the foraging dung beetle has been updated to: x i (t+1)=x i (t)+e -t / T ×(x i (t)-Lb b )+e -t / T ×(x i (t)-Ub b ); Among them, Lb b and Ub b Let Lb represent the upper and lower bounds of the optimal foraging region, respectively. b and Ub b The expressions are as follows: Among them, X b This is indicated as the optimal foraging location globally; Determine whether each dung beetle has exceeded the boundary; Update the current optimal solution and its fitness value; Repeat the above steps until the set number of iterations is reached, and output the global optimal solution of the parameters in the late LSTM model.

5. The method for late arrival reminders for ride-hailing orders according to claim 1, characterized in that, The steps for predicting late arrival times using the late arrival LSTM model include: The order information set was divided into an order training set and an order test set in a 7:3 ratio. Initialize the late LSTM model; The order features from the order training set are input into the late arrival LSTM model, and the parameters in the late arrival LSTM model are optimized using the dung beetle optimization algorithm to output a single predicted value for late arrival time. Calculate the lateness loss function between the predicted value of the lateness time and the actual value of the lateness time; If the late loss function converges, output the trained late LSTM model; otherwise, continue training. The expression for the late arrival loss function is: Where m1 represents the number of late orders in the training set; C i This represents the actual late time of the i-th order in the late training set; Let η represent the predicted late time of the i-th order in the late training set; η > 0 is the smoothing term.

6. The method for late arrival reminders for ride-hailing orders according to claim 1, characterized in that, The order lateness information includes the estimated lateness time and probability of lateness: The probability of lateness is calculated based on the driver's historical late orders, and the expression is: Where P is the probability of being late, and D c D represents the number of historical late orders for the driver. z This represents the total number of orders the driver has ever placed.

7. The method for late arrival reminders for ride-hailing orders according to claim 1, characterized in that, Determining whether the driver will be late based on the predicted lateness time also includes: When the prediction result is late, the number of orders dispatched to the driver is adjusted according to the lateness time and the lateness probability, as specifically expressed by: Where f∈(0,1) represents the adjustment coefficient, and PDL is the number of orders dispatched by the driver before the adjustment. new This represents the number of orders dispatched to drivers after the adjustment, T. l This represents the set threshold.

8. A late arrival reminder system for ride-hailing orders, characterized in that, include: Order Record Unit: Used to record the historical order information of ride-hailing drivers and build a ride-hailing order information set; Filtering unit: used to filter the order feature information to obtain order features; Prediction unit: Used to predict the lateness time of ride-hailing drivers; The discrimination unit is used to determine whether the driver is late based on the predicted lateness time. Output unit: Used to output late order information based on the discrimination result of the discrimination unit.

9. A late arrival reminder system for ride-hailing orders according to claim 8, characterized in that, The prediction unit includes: The order features are input into multiple late time prediction models to obtain multiple late time prediction values. Based on multiple individual prediction values ​​of lateness time, a lateness time prediction model is used to obtain the predicted value of lateness time. The late-arriving variable weight combination prediction model is constructed based on the induced ordered weighted multiple average operator; The expression for the induced ordered weighted multiple average operator is: in, Represented as the predicted late time based on the induced ordered weighted multiple average operator; α1, α2, ..., α n Let η1, η2, ..., η be the individual predicted values ​​of late arrival time from n late arrival time prediction models; n ω represents the prediction accuracy of n individual models for predicting lateness; k Let α represent the weighting coefficient assigned to the k-th lateness prediction single model; η-index(k) It is represented as the late time prediction value of the single model corresponding to the kth prediction accuracy after the prediction accuracy is arranged in ascending order; The expression for the late-arriving variable weight combination prediction model is: Where C represents the actual lateness time.

10. A late arrival reminder system for ride-hailing orders according to claim 9, characterized in that, The single model for predicting late arrival time includes a late arrival LSTM model: The parameters in the late LSTM model are optimized using an improved dung beetle optimization algorithm. The parameters include the number of hidden units, the learning rate, and the batch size; The specific steps include: Initialize a dung beetle swarm, where each dung beetle position in the swarm represents a set of parameter values; The prediction accuracy between the predicted late time output by the late LSTM model and the actual late time is calculated based on the location of each dung beetle, and the inverse of the prediction accuracy is used as the fitness value of each dung beetle. The dung beetle colony was divided into five subgroups in equal proportions: rolling, dancing, breeding, foraging, and stealing. The positions of all dung beetles are adaptively updated based on their fitness values. The breeding dung beetles reproduce and lay eggs in a specific area, which is: Among them, X * This represents the current local optimum; Lb * and Ub * Let Lb and Ub represent the lower and upper bounds of the spawning zone, respectively; T represents the maximum number of iterations; Lb and Ub represent the lower and upper bounds of the optimization problem, respectively; the formula for updating the position of the spawn ball is: B i (t+1)=X * +b1×(B i (t)-Lb * )+b2×(B i (t)-Ub * ); Among them, B i (t) represents the position information of the i-th egg in the t-th iteration, and b1 and b2 represent two independent 1×3 random vectors; The location of the foraging dung beetle has been updated to: x i (t+1)=x i (t)+e -t / T ×(x i (t)-Lb b )+e -t / T ×(x i (t)-Ub b ); Among them, Lb b and Ub b Let Lb represent the upper and lower bounds of the optimal foraging region, respectively. b and Ub b The expressions are as follows: Among them, X b This is indicated as the optimal foraging location globally; Determine whether each dung beetle has exceeded the boundary; Update the current optimal solution and its fitness value; Repeat the above steps until the set number of iterations is reached, and output the global optimal solution of the parameters in the late LSTM model.

Images