Method for improving driver experience in scene that passenger modifies destination

Abstract

The invention discloses a method for improving driver experience in a passenger destination modification scene, and the method remarkably improves the flexibility and efficiency in a driver and passenger matching process by introducing an optimization mechanism for passenger destination modification and combining the order receiving willingness verification and dynamic recommendation functions of a driver. On one hand, the method supports the dynamic adjustment of the pre-driving stage and the in-driving stage by relaxing the limitation of destination modification, and meanwhile, when the system judges the modification request, the rigid rule and the order receiving preference of the driver are comprehensively considered, so that the bad experience caused by violating the intention of the driver is avoided; and on the other hand, when the destination modification may be rejected, the system can intelligently analyze and recommend an alternative destination, and balance the passenger demand and the driver's willingness, thereby reducing the empty driving rate and complaint problem, and comprehensively improving the travel experience and service satisfaction of the driver and the passengers.

Description

Technical Field

[0001] The present invention belongs to the field of online car-hailing, and specifically, relates to a method for improving driver experience in a scenario where a passenger modifies a destination. Background Art

[0002] With the popularization of mobile Internet and the rise of the sharing economy, the online car-hailing industry has grown rapidly under the support of market supervision and government policies, gradually solving many problems in the traditional taxi industry and providing people with a convenient, economical and high-quality travel option.

[0003] During the service process, drivers usually have some personalized order requirements, such as hoping that the end point of the order is on the way to their desired destination, or hoping that the pick-up point is located in a specific area such as an airport or train station. To this end, drivers can adjust their order preferences by modifying the mode settings.

[0004] However, when passengers modify their destinations during a trip, the new destination may not match the driver's willingness to accept the order, resulting in a decline in the driver's service experience and even complaints. In addition, existing technical scenarios usually only support passengers to modify their destinations during a trip. If the driver and passengers fail to reach an agreement on the modification, the driver may drive empty or the passenger's needs may not be met, thus affecting the experience of both the driver and the passenger.

[0005] Currently, the system only determines whether to allow modification by judging whether the new destination meets the hard rules, and fails to fully consider the driver's willingness to accept the order. This approach not only affects the driver's experience, but also easily leads to complaints.

[0006] In view of this, the present invention is proposed. Summary of the invention

[0007] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a method for improving the driver experience in a scenario where a passenger modifies the destination, thereby solving the problems raised in the above-mentioned background technology.

[0008] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0009] A method for improving driver experience in a scenario where a passenger modifies a destination comprises the following steps:

[0010] Receive the destination modification request from the passenger, bind the request with the current order information, record the time and location of the modification request, and enter the rule verification phase. Then, perform hard rule verification on the new destination proposed by the passenger. If the hard rule fails, the request is directly rejected and the specific reason for failure is fed back to the passenger. After the hard rule verification passes, perform path optimization analysis on the new destination and calculate the dynamic impact of the modification on the current itinerary.

[0011] Based on the results of map navigation optimization, the new destination is matched with the driver's intention. If the driver's intention verification fails, the next step is entered; if it passes, the destination is allowed to be modified and the order information is directly updated;

[0012] For modification requests that do not meet the driver's wishes, the probability of the driver accepting the new destination is predicted. If it is predicted that the driver can accept the destination modification request, a confirmation request is sent to the driver. If the prediction shows that the driver's acceptance probability is low, the alternative destination recommendation process is directly entered.

[0013] If the system predicts that the driver will accept the destination modification request, a confirmation request will be sent to the driver, and the itinerary impact information will be displayed. The driver will choose whether to accept the modification request based on the information provided. If the driver confirms acceptance, the destination will be modified and the order information will be updated. If the driver refuses, the next step will be taken.

[0014] If the driver refuses the request to modify the destination, an alternative destination is recommended to the passenger based on the recommended destination. If the passenger accepts the recommended alternative destination, the order information is updated and the driver is notified. If the passenger refuses the alternative, the modification process is terminated.

[0015] Optionally, the steps for performing route optimization analysis on the new destination and calculating the dynamic impact of the modification on the current itinerary are:

[0016] Get the current vehicle location L through the map navigation function current 、Original destination D current 、New Destination D new The optimal path information, including driving time, driving distance and other data;

[0017] Get the estimated travel time T from the current location to the new destination through the map navigation function new , and the estimated travel time T from the current location to the original destination current , the expression of the change in driving time ΔT is: ΔT = T new -T current If ΔT>0, it means that the modified destination will increase the driving time. If ΔT<0, it means that the modified destination will shorten the driving time.

[0018] Calculate the driving distance S from the current location to the new destination new , and the driving distance S from the current location to the original destination current , the formula for the change in driving distance ΔS is: ΔS = S new -S current If ΔS>0, it means that the new destination will increase the driving distance; if ΔS<0, it means that the new destination will shorten the driving distance

[0019] Based on the changes in distance and time, combined with the platform pricing rules, the driver's revenue change ΔR and cost change ΔC caused by changing the destination are calculated, and the ratio of revenue to cost B is further calculated: If B>1, the modified destination is favorable to the driver; if B<1, the modified destination may be unfavorable to the driver;

[0020] If the driver turns on the on-the-go mode, analyze whether the new route deviates from the driver's on-the-go area A based on the driver's on-the-go settings preferred ,The degree of deviation of the new route from the en-route area is calculated through navigation: if the deviation value exceeds the driver’s preset deviation threshold, it is regarded as a route deviation, and the result is recorded and used as a reference for the recommendation of an alternative destination.

[0021] Optionally, the probability of a driver receiving a new destination is predicted by a machine learning algorithm based on the driver's historical behavior data as follows:

[0022] The time series data X = [x 1 , x 2 , ..., x N ] is input into the bidirectional LSTM model to extract the features of the forward (historical) and reverse (future) sequences respectively. The forward LSTM is responsible for capturing the temporal dependencies in the data related to past behaviors, and the reverse LSTM extracts the associated information of future data. Finally, the output of BiLSTM is expressed as

[0023] The time series features O extracted by BiLSTM t Input attention mechanism, by calculating the importance weight of each time step feature, to enhance the focus on key features. The attention mechanism first uses the formula Calculate the weight α i , and then perform weighted summation on the output of BiLSTM according to the weights to generate the global feature representation AH, which is expressed as:

[0024] The global historical features AH generated by the attention mechanism are combined with the current order features X current The complete input feature representation F is formed by splicing, where AH reflects the comprehensive characteristics of the driver's historical behavior, X current Contains real-time information about current orders;

[0025] The complete feature representation F is input into the fully connected layer, the feature is transformed through the nonlinear activation function, and then the probability P of the driver accepting the new destination is output through the Sigmoid function. accept , the Sigmoid function limits the model output to the range of [0,1]. If P accept >Thresholdaccept If the system determines that the driver is very likely to accept the request; otherwise, the system determines that the driver may reject the request;

[0026] According to the predicted acceptance probability P accept decide the subsequent processing logic. If the driver acceptance probability is high, send a confirmation request to the driver and provide itinerary impact information related to the new destination. If the driver acceptance probability is low, the system directly enters the alternative destination recommendation process to avoid wasting the driver's time.

[0027] Optionally, after inputting the time series data X = [x 1 , x 2 ,..., x N into the bidirectional LSTM model, the bidirectional LSTM model processes the time series data through the steps of the forget gate, input gate, cell state update, and output gate. The steps are as follows:

[0028] At time step t, calculate the forget ratio f t using the current input data x t-1 and the previous hidden state h t . Its expression is: f t =σ(W f [h t-1 , x t +b f ), where f t is the output of the forget gate, W f is the weight matrix of the forget gate, h t-1 is the cell state of the previous moment, x t is the input of the current moment, b f is the bias term. Then, determine the cell state information to be updated at the current time step through the input gate;

[0029] Update the cell state C t according to the outputs of the forget gate and the input gate. Its expression is: C t =f t ·C t-1 +i t ·AC t , where C t represents the cell state of the current time step, f t ·C t-1 represents the retained information of the previous cell state after passing through the forget gate, represents the new information added by the input gate;

[0030] The output gate controls the output information of the current time step, specifically including the following calculation process: Use the Sigmoid function to calculate the output ratio O t =σ(Wo [h t-1 , x t + b o ) Transform the current cell state C t using the tanh activation function and combine with the output ratio C t to generate the current hidden state h t = O t · tanh(C t ) The current hidden state h t is both the output result of the current time step and the input information for the next time step;

[0031] The bidirectional LSTM model processes time series data through forward LSTM and backward LSTM units respectively to generate a forward hidden state sequence and a backward hidden state sequence The final output is the concatenation result O of the forward and backward hidden states t .

[0032] Optionally, the input gate controls the new information to be added at the current time step and generates a candidate cell state. First, calculate the input ratio i through the Sigmoid function t : i t = σ(W i [h t-1 , x t + b i ), then use the tanh activation function to generate a candidate state vector

[0033] Among them, W i and W c are the weight matrices of the input gate and the candidate state respectively, b i and b c are bias terms. The output i t of the input gate and the candidate state jointly determine the content of the new information at the current time step.

[0034] Optionally, the hard rule check includes verifying whether the new destination complies with the traffic restriction rules, whether it is located in the no-order area, and whether it is restricted by the control fence.

[0035] Optionally, during prediction, it is based on the driver's historical behavior data, including order receiving records, modified receiving records, and order completion situations, as well as the current order characteristics.

[0036] After adopting the above technical solutions, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below:

[0037] By introducing an optimization mechanism for modifying the passenger's destination and combining the driver's order acceptance willingness verification and dynamic recommendation function, the present invention significantly improves the flexibility and efficiency in the driver-passenger matching process. On the one hand, the present invention relaxes the restrictions on destination modification, supports dynamic adjustment during the pre-trip and in-trip stages, and comprehensively considers hard rules and the driver's order acceptance preferences when the system determines the modification request, avoiding bad experiences caused by violating the driver's will. On the other hand, when the destination modification may be rejected, the system can intelligently analyze and recommend alternative destinations, balancing the passenger's needs and the driver's will, thereby reducing the empty driving rate and complaint problems and comprehensively improving the travel experience and service satisfaction of both the driver and the passenger.

[0038] The following further describes in detail the specific implementation manners of the present invention with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The following drawings in the description are only some embodiments. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. In the

[0040] In the drawings:

[0041] Figure 1 It is a flowchart of the method for improving the driver experience.

[0042] It should be noted that these drawings and the textual description are not intended to limit the scope of the concept of the present invention in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. SPECIFIC IMPLEMENTATION MANNERS

[0043] Now, the present invention will be further described in detail with reference to the accompanying drawings.

[0044] Please refer to Figure 1 As shown, in this embodiment, a method for improving the driver experience in the scenario of a passenger modifying the destination is provided, including the following steps:

[0045] Receive a destination modification request from the passenger side, bind the request to the current order information, and record the time and location of the modification request, and enter the subsequent rule verification link;

[0046] Perform a hard rule verification on the new destination proposed by the passenger. If the hard rule fails, directly reject the request and feedback the specific failure reason to the passenger; if it passes, enter the next step. After the hard rule verification passes, perform a path optimization analysis on the new destination, calculate the dynamic impact on the current trip, specifically including changes in driving time, driving distance, and the benefit-cost ratio after route optimization;

[0047] Based on the results of map navigation optimization, the new destination is checked against the driver's intention, including: whether it meets the driver's set en route mode, whether it is located in the en route area or en route point set by the driver, and whether it meets the driver's order type preference (such as whether to accept airport or train station orders). If the driver's intention verification fails, proceed to the next step; if it passes, the destination is allowed to be modified and the order information is directly updated;

[0048] For modification requests that do not meet the driver's wishes, the system uses a machine learning algorithm to predict the probability of the driver accepting the new destination. If the system predicts that the driver can accept the destination modification request, it sends a confirmation request to the driver. If the prediction shows that the driver has a low probability of accepting the request, it directly enters the alternative destination recommendation process.

[0049] If the system predicts that the driver may accept the destination modification request, a confirmation request will be sent to the driver, and the trip impact information will be displayed, including changes in driving time, driving distance, and revenue costs. The driver will choose whether to accept the modification request based on the information provided. If the driver confirms acceptance, the destination will be modified and the order information will be updated; if the driver refuses, the next step will be taken;

[0050] If the driver refuses to change the destination, the system will dynamically recommend an alternative destination to the passenger based on the map navigation analysis and machine learning prediction results. The alternative destination must meet the following conditions: be as close to the original destination as possible, reduce the deviation from the driver's en route pattern after the modification, and increase the probability that the driver will accept the alternative destination. The system will display the recommended plan to the passenger;

[0051] If the passenger accepts the recommended alternative destination, the system updates the order information and notifies the driver; if the passenger rejects the alternative, the modification process is terminated.

[0052] Example: A passenger places an order through an online car-hailing platform, with the starting point being "city center of city A" and the destination being "train station of city A". After the driver accepts the order, the system matches successfully and starts driving towards the starting point. During the trip, the passenger suddenly requests to change the destination from "train station" to "suburban airport of city A". In this scenario, the system processes the destination modification through the optimization mechanism of the present invention.

[0053] The system first performs a hard rule check on the new destination "suburban airport", including whether it complies with the traffic restriction rules, whether it is within the area where orders are allowed to be issued, and whether it complies with the current platform's operating policies. After the check is passed, it proceeds to the next step.

[0054] The system detects the driver's order preferences and finds that the driver set the "on the way mode" when accepting the order, and the desired destination is the southern area of ​​City A, while the new destination "suburban airport" is beyond the driver's on the way area. Since the new destination does not meet the driver's wishes, the system further sends a confirmation request to the driver.

[0055] The system sends a request to the driver with information about the impact on the trip (such as an increase in driving time of 20 minutes, an increase in driving distance of 15 kilometers, and an increase in estimated earnings of 30 yuan). After viewing the prompt, the driver rejects the modification request.

[0056] By analyzing the passenger's needs and the driver's preference for places to go, the system recommends an alternative destination to the passenger, "Southern High-Speed ​​Railway Station in City A". This destination is close to the "Railway Station" and is also within the driver's area of ​​interest. The passenger accepts the recommendation, and the system updates the order information and notifies the driver.

[0057] In the end, the passenger's demand to change the destination was met. Although the passenger did not reach the initially proposed "suburban airport", the alternative destination "Southern High-speed Railway Station" met some of the passenger's needs (such as quick access to the high-speed railway).

[0058] The driver's wish to take a detour was respected and the route was not deviated due to changes in the destination.

[0059] Through optimization, the system avoids conflicts between passengers and drivers, reduces complaints and empty trips, and improves service satisfaction.

[0060] In this embodiment, the steps of performing route optimization analysis on the new destination and calculating the dynamic impact of the modification on the current trip are:

[0061] Get the current vehicle location L through the map navigation function current 、Original destination D current 、New Destination D new The optimal path information, including driving time, driving distance and other data;

[0062] Get the estimated travel time T from the current location to the new destination through the map navigation function new , and the estimated travel time T from the current location to the original destination current , the expression of the change in driving time ΔT is: ΔT = T new -T current If ΔT>0, it means that the modified destination will increase the driving time. If ΔT<0, it means that the modified destination will shorten the driving time.

[0063] Calculate the driving distance S from the current location to the new destination new , and the driving distance S from the current location to the original destination currentThe formula for the change in driving distance ΔS is: ΔS = S new -S current If ΔS>0, it means that the new destination will increase the driving distance; if ΔS<0, it means that the new destination will shorten the driving distance

[0064] Based on the changes in distance and time, combined with the platform pricing rules, the driver's revenue change ΔR and cost change ΔC caused by changing the destination are calculated, and the ratio of revenue to cost B is further calculated: If B>1, the modified destination is favorable to the driver; if B<1, the modified destination may be unfavorable to the driver;

[0065] If the driver turns on the on-the-go mode, analyze whether the new route deviates from the driver's on-the-go area A based on the driver's on-the-go settings preferred ,The degree of deviation of the new route from the en-route area is calculated through navigation: if the deviation value exceeds the driver’s preset deviation threshold, it is regarded as a route deviation, and the result is recorded and used as a reference for the recommendation of an alternative destination.

[0066] In this embodiment, the probability of a driver accepting a new destination is predicted by a machine learning algorithm, and the steps of predicting based on the driver's historical behavior data are as follows:

[0067] The time series data X = [x 1 , x 2 , ..., x N ] is input into the bidirectional LSTM model to extract the features of the forward (historical) and reverse (future) sequences respectively. The forward LSTM is responsible for capturing the temporal dependencies in the data related to past behaviors, and the reverse LSTM extracts the associated information of future data. Finally, the output of BiLSTM is expressed as

[0068] The time series features O extracted by BiLSTM t Input attention mechanism, by calculating the importance weight of each time step feature, to enhance the focus on key features. The attention mechanism first uses the formula Calculate the weight α i , and then perform weighted summation on the output of BiLSTM according to the weights to generate the global feature representation AH, which is expressed as:

[0069] The global historical features AH generated by the attention mechanism are combined with the current order features X current The complete input feature representation F is formed by splicing, where AH reflects the comprehensive characteristics of the driver's historical behavior, X current Contains real-time information of current orders (such as distance to destination, time and change in revenue, etc.);

[0070] The complete feature representation F is input into the fully connected layer, and the feature transformation is performed through the non-linear activation function (ReLU), and then the probability P that the driver accepts the new destination is output through the Sigmoid function. accept , the Sigmoid function restricts the model output within the range of [0,1], indicating the possibility that the driver accepts the request. If P accept > Threshold accept (set threshold, such as 0.7), the system judges that the driver is likely to accept the request; otherwise, the system judges that the driver may reject the request;

[0071] According to the predicted acceptance probability P accept , the subsequent processing logic is determined. If the driver acceptance probability is high (exceeding the threshold), a confirmation request is sent to the driver, and trip impact information related to the new destination (such as trip time and revenue change) is provided. If the driver acceptance probability is low (below the threshold), the system directly enters the alternative destination recommendation link to avoid wasting the driver's time.

[0072] In this embodiment, after the time series data X = [x 1 , x 2 ,..., x N is input into the bidirectional LSTM model, the bidirectional LSTM model processes the time series data through the steps of the forget gate, input gate, cell state update, and output gate. The steps are as follows:

[0073] At time step t, according to the current input data x t and the previous hidden state h t-1 to calculate the forgetting ratio f t , its expression is: f t = σ(W f [h t-1 , x t + b f ), where f t is the output of the forget gate, W f is the weight matrix of the forget gate, h t-1 is the cell state of the previous moment, x t is the input of the current moment, b f is the bias term. Then, the cell state information to be updated at the current time step is determined through the input gate;

[0074] According to the outputs of the forget gate and the input gate, the cell state C t is updated, and its expression is: C t = f t ·C t-1 + i t ·AC t , where C tRepresents the cell state at the current time step, f t ·C t-1 Represents the retained information after the forget gate for the cell state at the previous moment, Represents the new information added by the input gate;

[0075] The output gate controls the output information at the current time step, specifically including the following calculation process: Use the Sigmoid function to calculate the output ratio O t =σ(W o [h t-1 , x t +b o ) to transform the current cell state C t Use the tanh activation function for transformation, and combine with the output ratio C t , to generate the current hidden state h t =O t ·tanh(C t ) The current hidden state h t Is both the output result at the current time step and the input information for the next time step;

[0076] The bidirectional LSTM model processes time series data through forward LSTM and backward LSTM units respectively, generating a forward hidden state sequence And a backward hidden state sequence The final output is the concatenation result O of the forward and backward hidden states t ,

[0077] In this embodiment, the input gate controls the new information to be added at the current time step and generates a candidate cell state. First, calculate the input ratio i through the Sigmoid function t : i t =σ(W i [h t-1 , x t +b i ), then use the tanh activation function to generate a candidate state vector

[0078] Where, W i And W c Are the weight matrices of the input gate and the candidate state respectively, b i And b c Are bias terms. The output i of the input gate t And the candidate state Together determine the content of the new information at the current time step.

[0079] In this embodiment, the hard rule check includes verifying whether the new destination complies with traffic restriction rules, whether it is located in an area where orders are prohibited from being issued, and whether it is restricted by a control fence.

[0080] In this embodiment, during prediction, it is based on the driver's historical behavior data, including order acceptance records, modified reception records, and order completion status, as well as the current order characteristics.

[0081] Explanation of related terms

[0082] Online car-hailing: Online car-hailing is a travel service provided based on an Internet platform. It connects passengers and drivers through a mobile application, facilitating passengers to book vehicles and providing functions such as real-time positioning and fare calculation. Passengers can select the destination and reservation time through the mobile application, and the system will automatically assign nearby registered drivers to pick up and drop off passengers.

[0083] Multi-tenant: An online car-hailing service provider can be regarded as a tenant in the system. There are multiple tenants in the online car-hailing aggregation platform.

[0084] Can provide services.

[0085] Ride-sharing mode: The driver can modify whether to enable the ride-sharing mode through the driver-side mobile application and set the ride-sharing method. The ride-sharing method is divided into a ride-sharing area or a ride-sharing point.

[0086] Before the trip: After the driver and passenger are successfully matched, the process from the driver starting the trip to driving to the passenger pick-up point.

[0087] During the trip: After the driver and passenger are successfully matched, the process from the passenger getting on the vehicle to driving to the passenger drop-off point.

[0088] Experimental environment:

[0089] Urban environment: In the same city, select the same time period (for example, morning and evening rush hours) and order type (such as online car-hailing instant orders) to ensure environmental consistency.

[0090] Data source: 200 real orders involving destination modification are randomly assigned to the control group and the experimental group to avoid bias caused by order characteristics.

[0091] Control group solution: Adopt the existing technology, and the system only judges whether to allow destination modification through hard rule checks, without considering the driver's order acceptance willingness or alternative recommendation function.

[0092] Comparison of experimental processes

[0093] The two groups of solutions process passenger destination modification requests in the same environment, and their differences are as follows:

[0094]

[0095]

[0096] The experimental data still follow the above experimental group and control group designs, and the processing procedures are run in the same environment. The specific data are as follows:

[0097]

[0098] It can be seen from the comparison of the experimental data of the control group and the experimental group in the same environment that:

[0099] The optimization mechanism of the present invention significantly improves the passing rate of destination modification requests (from 60% to 85%) and the driver acceptance rate (from 55% to 80%), and effectively reduces the empty driving rate caused by modification failures (from 30% to 10%).

[0100] The alternative destination recommendation function significantly improves the passenger satisfaction (from 3.8 points to 4.6 points) and reduces the complaint rate (from 15% to 5%).

[0101] The experimental group's solution reduces the dissatisfaction of drivers caused by destination modification through driver willingness verification and acceptance probability prediction, and at the same time improves the experience of both drivers and passengers.

[0102] The present invention is not limited to the above embodiments. Anyone should know that structural changes made under the inspiration of the present invention, as long as they have the same or similar technical solutions as the present invention, fall within the protection scope of the present invention. The technologies, shapes, and structures not described in detail in the present invention are all well-known technologies.

Claims

1. A method for improving driver experience in a scenario where a passenger modifies a destination, characterized in that: The following steps are involved: Receive the destination modification request from the passenger, bind the request with the current order information, record the time and location of the modification request, and enter the rule verification phase. Then, perform hard rule verification on the new destination proposed by the passenger. If the hard rule fails, the request is directly rejected and the specific reason for failure is fed back to the passenger. After the hard rule verification passes, perform path optimization analysis on the new destination and calculate the dynamic impact of the modification on the current itinerary. Based on the results of map navigation optimization, the new destination is matched with the driver's intention. If the driver's intention verification fails, the next step is entered; if it passes, the destination is allowed to be modified and the order information is directly updated; For modification requests that do not meet the driver's wishes, the probability of the driver accepting the new destination is predicted. If it is predicted that the driver can accept the destination modification request, a confirmation request is sent to the driver. If the prediction shows that the driver's acceptance probability is low, the alternative destination recommendation process is directly entered; If the system predicts that the driver will accept the destination modification request, a confirmation request will be sent to the driver, and the itinerary impact information will be displayed. The driver will choose whether to accept the modification request based on the information provided. If the driver confirms acceptance, the destination will be modified and the order information will be updated; If the driver refuses, proceed to the next step; If the driver refuses the request to modify the destination, an alternative destination is recommended to the passenger based on the recommended destination. If the passenger accepts the recommended alternative destination, the order information is updated and the driver is notified. If the passenger refuses the alternative, the modification process is terminated.

2. A method for improving driver experience in a scenario where a passenger modifies a destination according to claim 1, characterized in that: The steps to perform route optimization analysis on the new destination and calculate the dynamic impact of the modification on the current trip are: Get the current vehicle location L through the map navigation function current 、Original destination D current 、New Destination D new The optimal path information, including driving time, driving distance and other data; Get the estimated travel time T from the current location to the new destination through the map navigation function new , and the estimated travel time T from the current location to the original destination current , the expression of the change in driving time ΔT is: ΔT = T new -T current If ΔT>0, it means that the modified destination will increase the driving time. If ΔT<0, it means that the modified destination will shorten the driving time. Calculate the driving distance S from the current location to the new destination new , and the driving distance S from the current location to the original destination current , the formula for the change in driving distance ΔS is: ΔS = S new -S current If ΔS>0, it means that the new destination will increase the driving distance; if ΔS<0, it means that the new destination will shorten the driving distance Based on the changes in distance and time, combined with the platform pricing rules, the driver's revenue change ΔR and cost change ΔC caused by changing the destination are calculated, and the ratio of revenue to cost B is further calculated: If B>1, the modified destination is favorable to the driver; if B<1, the modified destination may be unfavorable to the driver; If the driver turns on the on-the-go mode, analyze whether the new route deviates from the driver's on-the-go area A based on the driver's on-the-go settings preferred ,The degree of deviation of the new route from the en-route area is calculated through navigation: if the deviation value exceeds the driver’s preset deviation threshold, it is regarded as a route deviation, and the result is recorded and used as a reference for the recommendation of an alternative destination.

3. The method for improving driver experience in the scenario of passenger modifying destination according to claim 1, characterized in that: The machine learning algorithm is used to predict the probability of a driver accepting a new destination. The steps of this prediction based on the driver’s historical behavior data are: The time series data X = [x1, x2, ..., x N ] is input into the bidirectional LSTM model to extract the features of the forward (historical) and reverse (future) sequences respectively. The forward LSTM is responsible for capturing the temporal dependencies in the data related to past behaviors, and the reverse LSTM extracts the associated information of future data. Finally, the output of BiLSTM is expressed as The time series features O extracted by BiLSTM t Input attention mechanism, by calculating the importance weight of each time step feature, to enhance the focus on key features. The attention mechanism first uses the formula Calculate the weight α i , and then perform weighted summation on the output of BiLSTM according to the weights to generate the global feature representation AH, which is expressed as: The global historical features AH generated by the attention mechanism are combined with the current order features X current The complete input feature representation F is formed by splicing, where AH reflects the comprehensive characteristics of the driver's historical behavior, X current Contains real-time information about current orders; The complete feature representation F is input into the fully connected layer, the feature is transformed through the nonlinear activation function, and then the probability P of the driver accepting the new destination is output through the Sigmoid function. accept , the Sigmoid function limits the model output to the range of [0,1]. If P accept >Threshold accept , the system judges that the driver is likely to accept the request; otherwise, the system judges that the driver is likely to reject the request; According to the predicted acceptance probability P accept , determines the subsequent processing logic. If the driver's acceptance probability is high, a confirmation request is sent to the driver and the trip impact information related to the new destination is provided. If the driver's acceptance probability is low, the system directly enters the alternative destination recommendation stage to avoid wasting the driver's time.

4. The method for improving driver experience in the scenario of passenger modifying destination according to claim 1, characterized in that: The time series data X = [x1, x2, ..., x N ]After being input into the bidirectional LSTM model, the bidirectional LSTM model processes the time series data through the forget gate, input gate, cell state update, and output gate steps. The steps are as follows: At time step t, the forget gate is used according to the current input data x t and the previous hidden state h t-1 Calculate the forgetting ratio f t , its expression is: t =σ(W f [h t-1 , x t ]+b f ), where f t is the output of the forget gate, W f is the weight matrix of the forget gate, h t-1 is the unit state at the previous moment, x t is the input at the current moment, b f is the bias term. Then, the cell state information that needs to be updated in the current time step is determined through the input gate. According to the output of the forget gate and the input gate, the cell state C t Update, its expression is: C t =f t ·C t-1 +i t ·AC t , where C t represents the cell state at the current time step, f t ·C t-1 Represents the retained information of the cell state at the previous moment after passing through the forget gate, Represents the new information added by the input gate; The output gate controls the output information of the current time step, which includes the following calculation process: Use the Sigmoid function to calculate the output ratio Q t =σ(W o [h t-1 , x t ]+b o ) for the current cell state C t Use the tanh activation function for transformation and combine it with the output ratio C t , generate the current hidden state h t =O t tanh(C t ) Current hidden state h t It is both the output result of the current time step and the input information of the next time step; The bidirectional LSTM model processes time series data through forward LSTM and reverse LSTM units respectively to generate a forward hidden state sequence and the reverse hidden state sequence The final output is the concatenation of the forward and reverse hidden states O t .

5. The method for improving driver experience in the scenario of passenger modifying destination according to claim 1, characterized in that: The input gate controls the new information that needs to be added in the current time step and generates candidate cell states. First, the input ratio i is calculated by the Sigmoid function. t :i t =σ(W i [h t-1 , x t ]+b i ), then, the tanh activation function is used to generate the candidate state vector Among them, W i and W c are the weight matrices of the input gate and candidate state, respectively, and b i and b c is the bias term, the output i of the input gate t and candidate status Together they determine the content of the new information at the current time step.

6. The method for improving driver experience in the scenario of passenger modifying destination according to claim 1, characterized in that: Hard rule checks include verifying whether the new destination complies with traffic restrictions, is located in a prohibited area, and is restricted by control fences.

7. The method for improving driver experience in the scenario of passenger modifying destination according to claim 1, characterized in that: The prediction is based on the driver’s historical behavior data, including order acceptance records, modified acceptance records and order completion status, as well as current order characteristics.

Images