Method for improving order receiving rate of reserved orders based on saas

Abstract

The invention discloses a method for improving the order receiving rate of a reservation order based on saas, and the method can effectively improve the order receiving rate of the reservation order of an online car-hailing platform through the combination of demand prediction and abnormal point recognition technologies. An order demand is accurately predicted by using a hybrid machine learning model, and the matching degree of the reservation order and the order receiving willingness of a driver is improved by dynamically adjusting the conversion critical time of the reservation order and the real-time order; meanwhile, through a 3-sigma anomaly detection mechanism, the vehicle use peak period or the sudden demand change is identified in time, the price strategy and the scheduling rule are dynamically adjusted, and the platform resource allocation is optimized. According to the method, the order receiving enthusiasm of the driver is remarkably improved, the waiting time of the passenger is shortened, the overall order completion rate of the platform is improved, the user experience is improved, and the market competitiveness of the platform is enhanced.

Description

Technical Field

[0001] The present invention belongs to the field of online car-hailing, and specifically relates to a method for improving the acceptance rate of reserved orders based on SaaS. Background Art

[0002] With the continuous rise and rapid development of the online car-hailing industry, reserved orders, as an important form of online car-hailing services, have become one of the important means to meet the diverse travel needs of users due to their flexibility and convenience. Users can reserve a vehicle in advance according to their own travel plans, which not only avoids the difficulty of using a car during peak hours but also reduces the long waiting time due to the shortage of vehicles. This service mode provides users with a personalized and efficient travel experience and also enhances the service diversity of the online car-hailing platform.

[0003] However, with the continuous increase in the number of reserved orders, the platform is gradually facing new challenges. The acceptance rate of reserved orders, as an important indicator to measure the service efficiency of online car-hailing, shows a gradually decreasing trend. The main reason is that the matching degree between reserved orders and drivers' willingness to accept orders has decreased. Specifically, since reserved orders usually include clear pick-up time, destination, and vehicle type requirements, these restrictive conditions may not be completely consistent with the actual acceptance preferences of drivers (such as distance, time period income, etc.). In addition, factors such as peak hours, remote destinations, or uneven order distribution will further lead to difficulties in matching some reserved orders with suitable drivers.

[0004] This phenomenon not only has a negative impact on the travel experience of passengers, such as the reserved order being difficult to be accepted in time, resulting in the obstruction of passengers' trips, but also causes losses to the overall operation efficiency of the platform. The decrease in the acceptance rate means that the platform fails to make full use of driver resources, affecting the conversion rate of orders and the number of completed orders, thus possibly weakening the market competitiveness of the platform.

[0005] In view of this, the present invention is specifically proposed. Summary of the Invention

[0006] 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 acceptance rate of reserved orders based on SaaS, which solves the problems raised in the above background art.

[0007] To solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0008] A method for improving the acceptance rate of reserved orders based on SaaS, comprising the following steps:

[0009] Receive online car-hailing order data and transmit the data to the cloud platform for processing;

[0010] Build a hybrid machine learning framework for order demand prediction. This framework has an autoregressive integrated moving average (ARIMA) model and a backpropagation neural network, and optimizes the processing on the local space of data by dynamically selecting a regression algorithm to predict the order demand for a certain period in the future from historical order data;

[0011] According to the demand prediction result, automatically convert the reserved order into a real-time order, and dynamically adjust the critical time of order conversion under the influence of the peak demand prediction period, and initiate real-time order scheduling in advance;

[0012] Combine an outlier detection mechanism, and through the 3-sigma outlier detection criterion, identify the abnormal fluctuations in the order demand in real time, and automatically mark the outliers and trigger the scheduling optimization mechanism,

[0013] According to the outlier identification result and demand prediction, adjust the price strategy and scheduling rules, and adopt a dynamic pricing strategy during abnormal demand periods.

[0014] Optionally, obtain the time series order demand data X, perform a difference calculation on the time series data, and judge whether it is stationary. If the time series is stationary, perform ARIMA model modeling and prediction. The steps are as follows:

[0015] For a stationary time series, build an autoregressive integrated moving average model, and the formula is: where: y t is the current time series value; λ i , θ j are the weight coefficients of the AR and MA terms; ∈ t is the random error term;

[0016] Automatically determine the model parameters p and q through the Bayesian information criterion. The specific formula is: BIC(p * , q * ) = kln(n) - 2ln(l), where k is the number of model parameters, n is the number of samples, and l is the maximum likelihood value.

[0017] Optionally, for a non-stationary time series, build a neural network model based on backpropagation. The input of the model is the historical order demand feature vector X = (y t-1 , y t-2 ,...., y t-n ), and the output is the predicted value y t ; then, perform network calculations through the following formula: A = f(WY + b), where W is the weight matrix; Y is the input feature vector; b is the bias vector; f(z) is the activation function, and the sigmoid function is used.

[0018] Optionally, construct a neural network model based on backpropagation to output the predicted value y t The steps are as follows:

[0019] S1: Build a multi-layer feedforward neural network, which includes an input layer, at least one hidden layer, and an output layer; among them, the number of nodes in the input layer is the same as the dimension n of the feature vector; the hidden layer adopts a fully connected structure, and the activation function is set as the Sigmoid function f(z); the output layer contains a single node for outputting the predicted value y t ;

[0020] S2: Randomly initialize the weight matrix W [l] and the bias vector b [l] , where l represents the number of layers, and the weights and biases are updated through the training process;

[0021] S3: Forward propagation calculation: Perform forward propagation calculation through the input feature vector X, and calculate the activation value layer by layer and pass it on. The activation value calculation formula for the l-th layer is: A [l] = f(W [l] A [l-1] + b [l] ), where A [l-1] represents the activation value of the previous layer; W [l] represents the weight matrix of the l-th layer; b [l] represents the bias vector of the l-th layer. When calculating the predicted value y t , obtain the activation value of the output layer through forward propagation as the predicted value y t of the model, that is, y t = A [L] , where L represents the output layer;

[0022] S4: Calculate the loss function according to the predicted value y t and the actual value . The loss function uses the mean square error (MSE), and the formula is: where m is the number of samples.

[0023] S5: Optimize the loss function J by gradient descent, and update the weight matrix and bias vector through the backpropagation algorithm;

[0024] S6: Repeat steps S1 to S5 until the model converges. After training, input the new time series feature X' for prediction and output the corresponding predicted value y t .

[0025] Optionally, adopt a dynamic selection regression algorithm during the process of switching models, and the steps are as follows:

[0026] For the test data point t i, calculate its similarity with the training data points based on the K-nearest neighbor algorithm, and the formula is: According to the similarity, select K nearest neighbor data points to form a local space;

[0027] Within the local space, calculate the mean absolute error for the prediction results of the ARIMA and BPNN models respectively, and the formula is: Finally, according to the comparison results of the MAE, dynamically select the optimal model, and use the predicted value of this model as the final predicted value of the test data point t i of.

[0028] Optionally, according to the demand prediction results, automatically convert the reservation order into a real-time order, and under the influence of the demand prediction peak period, dynamically adjust the critical time for order conversion. The steps to initiate real-time order scheduling in advance are as follows:

[0029] According to the demand prediction results, determine the conversion strategy during the peak period, and dynamically adjust the critical time for converting the reservation order into a real-time order. Specifically, during the demand prediction peak period, advance the conversion critical time, calculate the time difference between the current time and the reservation time in the reservation order data, and determine whether the conversion condition is met. If the time difference is less than or equal to the dynamically adjusted critical time, mark the reservation order as a real-time order;

[0030] For the reservation orders that have been converted into real-time orders, initiate the real-time scheduling mechanism in advance, allocate drivers in the real-time order queue through a matching algorithm, and preferentially select idle drivers who are closer to the passengers and have a high vehicle type matching degree; at the same time, adjust the scheduling range according to the regional distribution during the demand prediction peak period, expand the driver recall radius, and combine price incentives to enhance the driver's willingness to accept orders, thereby optimizing the order completion rate during the peak period.

[0031] Optionally, combined with the outlier detection mechanism, through the 3-sigma outlier detection criterion, real-time identify abnormal fluctuations in order demand, such as sudden increases or decreases in demand, and automatically mark the outliers and trigger the scheduling optimization mechanism. Automatically increase driver scheduling during sudden demand increases and reduce driver scheduling during low demand. The steps to optimize platform resource allocation are as follows:

[0032] Obtain real-time order demand data and historical order demand data, and perform standardized processing on the data in combination with external environment data to construct a time series data set;

[0033] Based on the historical order demand data, calculate the mean μ and standard deviation σ of the time series data, and determine the normal range of order demand according to the 3-sigma detection criterion. The upper and lower limits of outlier detection are μ + 3σ and μ - 3σ respectively, and the data points outside this range are marked as outliers;

[0034] By performing point-by-point detection on real-time order demand data, if the current data point exceeds the normal range of 3-sigma, it is marked as an abnormal point; if the order demand exceeds μ + 3σ, it is identified as a sudden increase in demand; if it is lower than μ - 3σ, it is identified as a sudden decrease in demand, and the time, area, and change trend of the abnormal point are recorded as the triggering conditions for subsequent scheduling optimization;

[0035] Based on the classification results of the detected abnormal points, when there is a sudden increase in demand, the driver scheduling strategy in the target area is dynamically adjusted, specifically including increasing the driver distribution in the abnormal area, expanding the driver recall radius, and enhancing the driver's willingness to accept orders through an incentive mechanism; when there is a sudden decrease in demand, the driver allocation in the low-demand area is reduced, and the idle drivers are preferentially scheduled to the high-demand area, thereby realizing the dynamic optimal allocation of platform resources;

[0036] Monitor the execution effect of the scheduling optimization measures in real time, record key indicators such as the order completion rate and response time in the abnormal area, return the monitoring results and the abnormal point marking data to the abnormal detection mechanism, and update the statistical parameters of the order demand through feedback to further improve the efficiency and accuracy of abnormal point detection and scheduling optimization.

[0037] Optionally, according to the abnormal point recognition result and demand prediction, the steps to adjust the price strategy and scheduling rules, and temporarily adjust the price or provide rewards to drivers through a dynamic pricing strategy during abnormal demand periods to balance the supply and demand relationship and encourage drivers to accept orders are as follows:

[0038] According to the identified abnormal demand periods and areas, dynamically adjust the price strategy; during the periods and in the areas with a sudden increase in demand, temporarily raise the price and increase the dynamic price increase coefficient;

[0039] On the basis of the dynamic price adjustment, optimize the scheduling rules according to the abnormal point recognition result; during the periods or in the areas with a sudden increase in demand, expand the driver recall radius, preferentially schedule the idle drivers closer, and increase the distribution density of driver resources in the high-demand area; during the periods or in the areas with a sudden decrease in demand, reduce the driver allocation in the low-demand area and schedule the remaining driver resources to the high-demand area.

[0040] 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:

[0041] 1. By combining demand forecasting and outlier identification technologies, the present invention can effectively improve the acceptance rate of reserved orders on the online car-hailing platform. It uses a hybrid machine learning model to accurately predict order demand, and improves the matching degree between reserved orders and drivers' willingness to accept orders by dynamically adjusting the conversion critical time between reserved orders and real-time orders. At the same time, through a 3-sigma anomaly detection mechanism, it can timely identify peak car-using periods or sudden demand changes, dynamically adjust price strategies and dispatching rules, and optimize the platform's resource allocation. This method not only significantly improves drivers' enthusiasm for accepting orders, reduces passengers' waiting time, but also enhances the overall order completion rate of the platform, improves the user experience, and enhances the platform's market competitiveness.

[0042] The following further describes in detail the specific implementation manners of the present invention with reference to the accompanying drawings. Description of the Drawings

[0043] The accompanying drawings in the following 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 attached

[0044] In the figures:

[0045] Figure 1 It is a flowchart of a method for improving the acceptance rate of reserved orders.

[0046] It should be noted that these drawings and text descriptions 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 Embodiment

[0047] Now, the present invention will be further described in detail with reference to the accompanying drawings.

[0048] Please refer to Figure 1 As shown, in this embodiment, a method for improving the acceptance rate of reserved orders based on SaaS is provided, including the following steps:

[0049] Receive online car-hailing order data, including but not limited to the passenger's reservation time, departure location, destination, vehicle type requirements, driver's current location, weather, and traffic conditions, and transmit the data to the cloud platform for processing;

[0050] Construct a hybrid machine learning framework for order demand forecasting. This framework has an autoregressive integrated moving average model with differences and a backpropagation neural network, and optimizes the processing in the local space of the data by dynamically selecting a regression algorithm to predict the order demand for a certain period in the future from historical order data;

[0051] Automatically convert reservation orders into real-time orders according to the demand prediction results, and dynamically adjust the critical time for order conversion under the influence of the peak demand prediction period to initiate real-time order scheduling in advance;

[0052] Combined with the anomaly detection mechanism, through the 3-sigma anomaly detection criterion, real-time identify abnormal fluctuations in order demand, such as sudden increases or decreases in demand, and automatically mark the anomaly points and trigger the scheduling optimization mechanism. Automatically increase driver scheduling when demand surges and reduce driver scheduling during low demand to optimize platform resource allocation;

[0053] According to the anomaly point identification results and demand prediction, adjust the price strategy and scheduling rules. During abnormal demand periods (such as holidays, bad weather, etc.), through dynamic pricing strategies, temporarily adjust the price or provide incentives to drivers to balance the supply and demand relationship and encourage drivers to accept orders. After the driver accepts the order, a new content of the pick-up time reminder will be broadcast. The following is an example: New broadcast: The passenger needs a ride at 10:10 today. Please wait patiently.

[0054] The online car-hailing platform will receive passengers' order information in real time, such as the reserved pick-up time, departure and arrival locations, vehicle type requirements, etc. Combining with the driver's current location, weather, traffic and other conditions, these data are transmitted to the cloud for processing. Then, the platform uses a very intelligent algorithm framework to predict future order demand. This framework is "hybrid", which can not only analyze past data trends (such as the high or low demand during a certain period), but also discover some hidden rules through more complex calculations. The platform will also convert reservation orders into real-time orders in advance according to the predicted order demand, especially during peak periods, accelerate the processing of these orders and schedule drivers in advance.

[0055] If there are some unexpected situations, such as suddenly many people needing rides or a sudden decrease in ride-hailing demand, the system will detect these abnormal situations through a mechanism called "3-sigma". For example, if the orders in a certain area suddenly skyrocket, the platform will increase the number of drivers to handle it; conversely, if the demand is very low, reduce the scheduling of drivers in that area and transfer them to other places to make the drivers' time more efficient.

[0056] In addition, the platform will also adjust the price according to these demand predictions and abnormal situations. For example, the price will be slightly increased during peak periods to attract drivers to accept orders; during some off-peak periods or special weather, the platform will give extra incentives to drivers to encourage them to accept more orders.

[0057] Finally, when the driver accepts the passenger's order, the system will remind the driver through voice, such as: "The passenger reserved a ride at 10:10 today. Please wait patiently!" This not only allows the driver to know the passenger's pick-up time, but also improves the overall service experience.

[0058] A method for improving the acceptance rate of reservation orders based on SaaS, as described in claim 1, is characterized in that time series order demand data X is obtained, and the time series data is subjected to difference calculation to determine whether it is stationary. If the time series is stationary, ARIMA model modeling and prediction are performed, and the steps are as follows:

[0059] For a stationary time series, a differential integrated moving average autoregressive model is constructed, and the formula is: where: y t : the current time series value; λ i , θ j : the weight coefficients of the AR and MA terms; ∈ t : the random error term;

[0060] The parameters p and q of the model are automatically determined through the Bayesian Information Criterion (BIC). The specific formula is: BIC(p * , q * ) = kln(n) - 2ln(l), where k is the number of model parameters, n is the number of samples, and lll is the maximum likelihood value.

[0061] A method for improving the acceptance rate of reservation orders based on SaaS, as described in claim 1, is characterized in that for a non-stationary time series, a neural network model based on backpropagation is constructed. The input of the model is the historical order demand feature vector X = (y t-1 , y t-2 ,...., y t-n ), and the output is the predicted value y t ; network calculation is performed through the following formula: A = f(WY + b), where: W: weight matrix; Y: input feature vector; b: bias vector; f(z): activation function, using the sigmoid function.

[0062] A method for improving the acceptance rate of reservation orders based on SaaS, as described in claim 1, is characterized in that the steps for constructing a neural network model based on backpropagation to output the predicted value yt are as follows:

[0063] S1: Build a multi-layer feedforward neural network, which includes an input layer, at least one hidden layer, and an output layer; among them, the number of nodes in the input layer is the same as the dimension n of the feature vector; the hidden layer adopts a fully connected structure, and the activation function is set as the sigmoid function f(z); the output layer contains a single node for outputting the predicted value y t

[0064] S2: Randomly initialize the weight matrix W [l] and the bias vector b [l], where l represents the number of layers, and the weights and biases are updated through the training process.

[0065] S3: Forward propagation calculation: Perform forward propagation calculation through the input feature vector X, and calculate the activation values layer by layer. The formula for the activation value of the l-th layer is: A [l] = f(W [l] A [l-1] + b [l] ), where A [l-1] represents the activation value of the previous layer; W [l] represents the weight matrix of the l-th layer; b [l] represents the bias vector of the l-th layer. Calculate the predicted value y t through forward propagation, obtain the activation value of the output layer, and use it as the predicted value y t of the model, that is, y t = A [L] ), where L represents the output layer.

[0066] S4: Calculate the loss function based on the predicted value y t and the actual value . The loss function uses the mean squared error (MSE), and the formula is: where m is the number of samples.

[0067] S5: Backpropagation and parameter update: Optimize the loss function J using gradient descent, and update the weight matrix and bias vector through the backpropagation algorithm.

[0068] S6: Model training and prediction output: Repeat steps S1 to S5 until the model converges. After training, input the new time series feature X' for prediction and output the corresponding predicted value y t .

[0069] According to the method for improving the acceptance rate of reservation orders based on SaaS as claimed in claim 1, characterized in that, during the process of switching the model, a dynamic selection regression algorithm is adopted, and its steps are:

[0070] For the test data point t i , calculate its similarity with the training data points based on the K-nearest neighbor algorithm (KNN), and the formula is: According to the similarity, select K nearest neighbor data points to form a local space;

[0071] Within the local space, calculate the mean absolute error (MAE) for the prediction results of the ARIMA and BPNN models respectively, and the formula is: Finally, according to the comparison result of the MAE, dynamically select the optimal model, and use the predicted value of this model as the final predicted value of the test data point t i .

[0072] A method for improving the acceptance rate of reservation orders based on SaaS, as described in claim 1, is characterized in that according to the demand prediction result, the reservation order is automatically converted into a real-time order, and under the influence of the demand prediction peak period, the critical time for order conversion is dynamically adjusted. The steps for starting the real-time order scheduling in advance are as follows

[0073] According to the demand prediction result, determine the conversion strategy during the peak period, and dynamically adjust the critical time for converting the reservation order into a real-time order. Specifically, during the demand prediction peak period, the conversion critical time is advanced (for example, adjusted from the original set 30 minutes to 20 minutes). Calculate the time difference between the current time and the reservation time in the reservation order data, and determine whether the conversion condition is met. If the time difference is less than or equal to the dynamically adjusted critical time, then mark the reservation order as a real-time order;

[0074] For the reservation orders that have been converted into real-time orders, start the real-time scheduling mechanism in advance. Allocate drivers in the real-time order queue through a matching algorithm, and preferentially select idle drivers who are closer to the passengers and have a high vehicle type matching degree; at the same time, adjust the scheduling range according to the regional distribution during the demand prediction peak period, expand the driver recall radius, and combine price incentive measures (such as peak period rewards) to enhance the driver's willingness to accept orders, thereby optimizing the order completion rate during the peak period.

[0075] A method for improving the acceptance rate of reservation orders based on SaaS, as described in claim 1, is characterized in that in combination with an outlier detection mechanism, through the 3-sigma outlier detection criterion, the abnormal fluctuations in the order demand are identified in real time, such as sudden increases or decreases in demand, and the outlier points are automatically marked and the scheduling optimization mechanism is triggered. When the demand suddenly increases, the driver scheduling is automatically increased, and when the demand is low, the driver scheduling is reduced. The steps for optimizing the platform resource allocation are as follows

[0076] Obtain the real-time order demand data and historical order demand data, and perform standardization processing on the data in combination with external environment data (such as weather, traffic conditions, and holiday information) to construct a time series data set;

[0077] Based on the historical order demand data, calculate the mean μ and standard deviation σ of the time series data, and determine the normal range of the order demand according to the 3-sigma detection criterion. The upper and lower limits of the outlier detection are μ + 3σ and μ - 3σ respectively, and the data points outside this range are marked as outlier points;

[0078] By performing point-by-point detection on the real-time order demand data, if the current data point exceeds the 3-sigma normal range, then mark it as an outlier point; if the order demand exceeds μ + 3σ, it is identified as a sudden increase in demand; if it is lower than μ - 3σ, it is identified as a sudden decrease in demand, and record the time, region, and change trend of the outlier point as the trigger condition for subsequent scheduling optimization;

[0079] Based on the classification results of detected abnormal points, when there is a sudden increase in demand, dynamically adjust the driver dispatching strategy for the target area, specifically including increasing the driver distribution in the abnormal area, expanding the driver recall radius, and enhancing the driver's willingness to accept orders through incentive mechanisms (such as peak rewards or commission-free policies); when there is a sudden decrease in demand, reduce the driver allocation in low-demand areas and preferentially dispatch idle drivers to high-demand areas, thereby achieving the dynamic optimal allocation of platform resources;

[0080] Real-time monitor the execution effect of the dispatching optimization measures, record key indicators such as the order completion rate and response time in the abnormal area, return the monitoring results and the abnormal point marking data to the abnormal detection mechanism, and further improve the efficiency and accuracy of abnormal point detection and dispatching optimization by updating the statistical parameters of order demand through feedback.

[0081] A method for improving the acceptance rate of reserved orders based on SaaS as claimed in claim 1, characterized in that, according to the abnormal point recognition result and demand prediction, adjusting the price strategy and dispatching rules, and during abnormal demand periods (such as holidays, bad weather, etc.), through dynamic pricing strategies, temporarily adjusting the price or providing rewards to drivers to balance the supply and demand relationship and encourage drivers to accept orders, the steps are as follows:

[0082] According to the identified abnormal demand periods and regions, dynamically adjust the price strategy; in the regions and periods with a sudden increase in demand, temporarily raise the price and increase the dynamic price increase coefficient; in the regions and periods with a sudden decrease in demand, lower the price to stimulate passenger demand; for high-demand regions during peak hours, provide driver incentive policies such as peak rewards or commission-free to enhance the driver's willingness to accept orders and relieve the imbalance between supply and demand.

[0083] On the basis of dynamic price adjustment, optimize the dispatching rules according to the abnormal point recognition result; in the regions or periods with a sudden increase in demand, expand the driver recall radius, preferentially dispatch idle drivers closer in distance, and increase the distribution density of driver resources in high-demand areas; in the regions or periods with a sudden decrease in demand, reduce the driver allocation in low-demand areas and dispatch the remaining driver resources to areas with higher demand, realizing the dynamic optimal allocation of platform resources.

[0084] Reserved order: A reserved order refers to an order generated after a passenger successfully books a vehicle in advance through a reservation platform when using a online car-hailing service. This order includes information such as the passenger's departure location, destination, reservation time, vehicle type, etc. The driver can accept the order according to his own will. If it does not meet his own will, he can refuse to accept it. After accepting the order, the driver shall complete the service according to the information on the reserved order.

[0085] Acceptance rate: The acceptance rate is an indicator in the online car-hailing industry, which refers to the ratio between the number of orders received by a driver within a certain period of time and the number of orders dispatched by the platform

[0086] Experimental objective: To verify whether the present invention (the reservation order optimization method combining demand prediction and outlier detection) can significantly improve the reservation order acceptance rate and passenger experience under the same environment and configuration.

[0087] Experimental conditions:

[0088] To verify the effectiveness of the present invention, the same driver configuration conditions were strictly set in the experiment. Specifically: 1000 drivers were respectively configured in the experimental group and the control group, including 700 economy cars, 200 comfortable cars, and 100 luxury cars. The service scores of all drivers were above 4.5 points, and the average daily online time was 10 hours; the participation rate of online drivers during peak hours (7:00 - 9:00 in the morning and 17:00 - 19:00 in the evening) was 90%. The initial positions of the drivers were evenly distributed within the urban area and were adjusted to the high-demand areas predicted by the demand prediction according to the dispatching strategy during peak hours. Through the driver configuration with the same quantity and conditions, it was ensured that the experimental data could accurately reflect the true differences between the present invention and the traditional method in the optimization of the reservation order acceptance rate.

[0089] Experimental group setting:

[0090] Experimental group: Apply the method of the present invention, combining demand prediction, outlier detection, dynamic price strategy, and dispatching optimization.

[0091] Control group: Adopt the traditional dispatching method, which only matches based on the order submission time and the driver's current position, without prediction and dynamic optimization.

[0092] Experimental indicators:

[0093] Reservation order acceptance rate: The percentage of the number of successfully accepted reservation orders in the total number of reservation orders. Reservation order response time: The average time from when the passenger submits a reservation to when the driver accepts the order. Reservation order completion rate: The percentage of the number of successfully completed reservation orders in the number of successfully accepted reservation orders. Peak-hour order backlog rate: The percentage of unaccepted reservation orders during peak hours in the total number of reservation orders.

[0094]

[0095] The experimental group identified the demand in peak areas in advance through demand prediction, combined with dynamic dispatching strategies and incentives, and the acceptance rate increased by 14 percentage points compared with the control group. The experimental group reduced the average response time from 5.8 minutes in the control group to 2.5 minutes, a reduction of 57% through dynamic driver allocation and preferential matching of highly willing drivers. The experimental group reduced the order cancellations caused by poor matching between drivers and orders, and the completion rate increased by 12 percentage points. The experimental group used the outlier detection mechanism to dispatch drivers in a timely manner during peak hours, and the number of unaccepted reservation orders during peak hours decreased significantly.

[0096] The present invention is not limited to the above embodiments. Any person 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, all 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 the acceptance rate of reservation orders based on SaaS, characterized in that: The following steps are involved: Receive online ride-hailing order data and transmit the data to the cloud platform for processing; Construct a hybrid machine learning framework for order demand forecasting. The framework has a differential integrated moving average autoregressive model and a back-propagation neural network. It also optimizes the local space of the data by dynamically selecting the regression algorithm to predict the order demand in a certain period of time in the future from the historical order data. According to the demand forecast results, the scheduled orders are automatically converted into real-time orders. Under the influence of the demand forecast peak period, the critical time of order conversion is dynamically adjusted to start the real-time order scheduling in advance. Combined with the outlier detection mechanism, the 3-sigma anomaly detection criterion is used to identify abnormal fluctuations in order demand in real time, automatically mark the anomalies and trigger the scheduling optimization mechanism; According to the results of outlier identification and demand forecast, the price strategy and scheduling rules are adjusted, and dynamic pricing strategies are implemented during periods of abnormal demand.

2. According to the method of claim 1, which is characterized by: Obtain the time series order demand data X, perform differential calculation on the time series data to determine whether it is stable. If the time series is stable, perform ARIMA model building and prediction. The steps are as follows: For the stationary time series, a differential integrated moving average autoregressive model is constructed, and the formula is: Where: y t is the current time series value; i ,θ j is the weight coefficient of AR and MA terms; ∈ t is the random error term; The model parameters p and q are automatically determined by the Bayesian Information Criterion. The specific formula is: BIC(p * ,q * )=kln(n)-2ln(l), where k is the number of model parameters, n is the number of samples, and lll is the maximum likelihood value.

3. The method for improving the acceptance rate of reservation orders based on SaaS according to claim 1 is characterized in that: For non-stationary time series, a neural network model based on back propagation is constructed. The input of the model is the historical order demand feature vector X = (y t-1 ,y t-2 , ..., y t-n ), the output is the predicted value y t ; Then, the network calculation is performed using the following formula: A = f(WY+b), Among them, W is the weight matrix; Y is the input feature vector; b is the bias vector; f(z) is the activation function, which uses the sigmoid function.

4. The method for improving the acceptance rate of reservation orders based on SaaS according to claim 1 is characterized in that: Construct a neural network model based on back propagation to output the predicted value y t The steps are: S1: Build a multi-layer feedforward neural network, which includes an input layer, at least one hidden layer and an output layer. The number of nodes in the input layer is the same as the dimension n of the feature vector. The hidden layer adopts a fully connected structure and the activation function is set to the Sigmoid function f(z). The output layer contains a single node for outputting the predicted value y. t ; S2: Randomly initialize the weight matrix W of the neural network [l] and the bias vector b [l] , where l represents the number of layers, weights and biases are updated through the training process; S3: Forward propagation calculation: forward propagation calculation is performed by inputting feature vector X, passing it layer by layer and calculating the activation value of each layer. The activation value calculation formula of the lth layer is: A [l] =f(W [l] A [l-1] +b [l] ), where A [l-1] Represents the activation value of the previous layer; W [l] represents the weight matrix of the lth layer; b [l] Represents the bias vector of the lth layer. Calculate the predicted value y t Through forward propagation, the activation value of the output layer is obtained as the predicted value y of the model t , that is, y t =A [L] , where L represents the output layer; S4: According to the predicted value y t and actual value Calculate the loss function. The loss function uses the mean square error (MSE) and the formula is: Among them, m is the number of samples. S5: Perform gradient descent optimization on the loss function J and update the weight matrix and bias vector through the back propagation algorithm; S6: Repeat steps S1 to S5 until the model converges. After training is completed, input the new time series feature X′ for prediction and output the corresponding predicted value y t .

5. The method for improving the acceptance rate of reservation orders based on SaaS according to claim 1 is characterized in that: In the process of switching models, a dynamic selection regression algorithm is used, and the steps are as follows: For the test data point t i , based on the K nearest neighbor algorithm, the similarity between it and the training data point is calculated. The formula is: According to the similarity, K nearest neighbor data points are selected to form the local space; In the local space, the mean absolute error of the prediction results of the ARIMA and BPNN models is calculated respectively, and the formula is: Finally, based on the comparison results of MAE, the optimal model is dynamically selected and the predicted value of the model is used as the test data point t i The final predicted value.

6. The method for improving the acceptance rate of reservation orders based on SaaS according to claim 1 is characterized in that: According to the demand forecast results, the scheduled orders are automatically converted into real-time orders, and the critical time of order conversion is dynamically adjusted under the influence of the demand forecast peak period. The steps to start real-time order scheduling in advance are as follows: According to the demand forecast results, determine the conversion strategy during the peak period, and dynamically adjust the critical time for converting the scheduled order to a real-time order. Specifically, during the peak demand forecast period, advance the critical time for conversion, calculate the time difference between the current time and the scheduled time in the scheduled order data, and determine whether the conversion conditions are met. If the time difference is less than or equal to the dynamically adjusted critical time, mark the scheduled order as a real-time order. For the reservation orders that have been converted into real-time orders, the real-time dispatch mechanism is started in advance, and drivers are assigned in the real-time order queue through the matching algorithm, with priority given to idle drivers who are close to the passengers and have a high degree of vehicle model matching; At the same time, the dispatch range is adjusted according to the regional distribution of demand forecast during peak hours, the driver recall radius is expanded, and price incentives are combined to increase drivers' willingness to accept orders, thereby optimizing the order completion rate during peak hours.

7. The method for improving the acceptance rate of reservation orders based on SaaS according to claim 1 is characterized in that: Combined with the anomaly detection mechanism, through the 3-sigma anomaly detection criterion, abnormal fluctuations in order demand, such as sudden increase or decrease in demand, are identified in real time, and the anomaly points are automatically marked and the dispatch optimization mechanism is triggered. Driver dispatch is automatically increased when demand increases suddenly, and driver dispatch is reduced when demand is low. The steps to optimize platform resource allocation are: Obtain real-time order demand data and historical order demand data, standardize the data in combination with external environment data, and construct a time series data set; Based on the historical order demand data, the mean μ and standard deviation σ of the time series data are calculated, and the normal range of order demand is determined according to the 3-sigma detection criterion, where the upper and lower limits of abnormal detection are μ+3σ and μ-3σ respectively. Data points outside this range are marked as abnormal points. By testing the real-time order demand data point by point, if the current data point exceeds the normal range of 3-sigma, it will be marked as an abnormal point; if the order demand exceeds μ+3σ, it will be identified as a sudden increase in demand; if it is lower than μ-3σ, it will be identified as a sudden drop in demand, and the time, area and change trend of the abnormal point will be recorded as the trigger condition for subsequent scheduling optimization; Based on the classification results of the detected anomalies, when demand suddenly increases, the driver dispatch strategy in the target area is dynamically adjusted, including increasing the distribution of drivers in the abnormal area, expanding the driver recall radius, and improving the driver's willingness to accept orders through incentive mechanisms; when demand suddenly drops, the driver allocation in the low-demand area is reduced, and idle drivers are dispatched to the high-demand area first, so as to achieve dynamic optimization of platform resources. Monitor the execution effect of scheduling optimization measures in real time, record key indicators such as order completion rate and response time in abnormal areas, feed the monitoring results and abnormal point marking data back to the anomaly detection mechanism, and update the statistical parameters of order requirements through feedback, so as to further improve the efficiency and accuracy of anomaly point detection and scheduling optimization.

8. The method for improving the acceptance rate of reservation orders based on SaaS according to claim 1 is characterized in that: According to the abnormal point identification results and demand forecast, adjust the price strategy and dispatch rules. Use dynamic pricing strategy during abnormal demand periods to temporarily adjust prices or provide incentives to drivers to balance supply and demand and motivate drivers to accept orders. The steps are as follows: Dynamically adjust pricing strategies based on the identified periods and areas of abnormal demand; temporarily increase prices and increase dynamic markup coefficients in areas and periods of sudden demand; Based on dynamic price adjustments, optimize dispatch rules according to the results of abnormal point identification; in areas or time periods with sudden increases in demand, expand the driver recall radius, give priority to dispatching idle drivers who are closer, and increase the distribution density of driver resources in high-demand areas; In areas or periods where demand drops suddenly, reduce the number of drivers allocated to low-demand areas and dispatch the remaining driver resources to areas with high demand.

Images