Online car-hailing supervision method and system

Abstract

The invention relates to the technical field of online car-hailing supervision, in particular to an online car-hailing supervision method and system. Comprising the steps of collecting data and performing data preprocessing to obtain an initial data set; inputting the initial data set into an urban online car-hailing saturation judgment model; outputting a judgment result through a data input layer, a data prediction layer, a data training layer and a judgment output layer in sequence; the judgment result is used for representing whether the online car-hailing number of the city is saturated or not; the data prediction layer predicts the initial data set through an RVM prediction model and an LSTM prediction model to obtain a prediction data set; the data training layer comprises a trained BP neural network; and the BP neural network carries out training on the prediction data set and outputs a training result. The invention provides an urban online car-hailing saturation judgment model which can perform prediction according to historical data of a city and judge whether the number of urban online car-hailing cars is saturated or not.

Description

Technical Field

[0001] This invention relates to the field of ride-hailing supervision technology, specifically to a ride-hailing supervision method and system. Background Technology

[0002] Ride-hailing services refer to online booking of taxis. These services utilize internet technology to build a platform that integrates supply and demand, breaking down information barriers between vehicles, drivers, and customers, thus supplementing the traditional taxi model. Their flexible working hours, relatively high monthly income, and low entry barriers have attracted a large number of people. However, with the increase in the number of ride-hailing vehicles, some cities have reached saturation, leading to a sharp decline in driver income, increased forced carpooling, and illegal activities, placing immense pressure on ride-hailing platforms. Therefore, effective methods for regulating the number of ride-hailing vehicles are needed to determine and predict whether a specific city's ride-hailing fleet is saturated, enabling ride-hailing platforms to manage and adjust accordingly.

[0003] Existing research methods mainly focus on short-term ride-hailing demand within cities. Patent CN202111202355.5 provides a method for predicting the spatiotemporal heat of ride-hailing demand based on a deep aggregation neural network. This method trains and optimizes the network model by acquiring weather and POI data for urban areas, thereby predicting the city's ride-hailing demand. Patent CN202210038032.5 provides a ride-hailing demand prediction method based on a spatiotemporal multi-graph neural network. This method constructs a spatiotemporal graph convolutional neural network; then, the order time and matched road segment are input into the spatiotemporal graph convolutional neural network to obtain the predicted ride-hailing demand value for that road segment.

[0004] The above methods are effective for short-term demand testing in urban areas, but they are inevitably inadequate when applied to the entire city. This is because the data and methods used are mostly at the micro level, the predicted time intervals are relatively short, and the judgment of the saturation of ride-hailing services in the entire city is not accurate enough.

[0005] Based on the above factors, the present invention provides a method and system for regulating ride-hailing services. Summary of the Invention

[0006] To better understand whether the demand for ride-hailing services in a city is saturated, and to help ride-hailing platforms make strategic arrangements, this invention provides a ride-hailing supervision method and system. It collects historical urban data and constructs an urban ride-hailing saturation index. Addressing the issue of insufficient data sample size, this invention uses a weighted k-fold cross-validation method for model training to obtain an urban ride-hailing saturation judgment model. This model is then used to identify predicted data for the city and determine whether the number of ride-hailing vehicles in the city is approaching saturation in the future.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for regulating ride-hailing services, characterized by comprising:

[0009] Historical data at the city level was collected and preprocessed to obtain an initial dataset;

[0010] The initial dataset is input into the urban ride-hailing saturation judgment model; the initial dataset passes through the data input layer, data prediction layer, data training layer and judgment output layer of the urban ride-hailing saturation judgment model in sequence, and outputs the judgment result; the judgment result is used to indicate whether the number of ride-hailing vehicles in the city is saturated;

[0011] The data input layer is used to input the initial dataset into the urban ride-hailing saturation judgment model;

[0012] The data prediction layer is trained and predicted based on the initial dataset to obtain a prediction dataset; the data prediction layer includes an RVM prediction model and an LSTM prediction model; the RVM prediction model is used to predict the initial dataset to obtain an RVM prediction dataset; the LSTM prediction model is used to predict the initial dataset to obtain an LSTM prediction dataset; the RVM prediction dataset and the LSTM prediction dataset are weighted to obtain the prediction dataset.

[0013] The data training layer includes a trained BP neural network; the BP neural network trains the prediction dataset and outputs the training results; the BP neural network includes an improved cuckoo optimization algorithm for model optimization; the judgment output layer makes a judgment based on the training results and outputs the judgment results.

[0014] The initial dataset includes a first type of data and a second type of data;

[0015] The first category of data includes urban resident population, urban GDP per capita, urban traffic mileage, urban car ownership per capita, and urban weather index; the second category of data includes the number of ride-hailing vehicles in the city.

[0016] The prediction dataset contains a first type of prediction data and a second type of prediction data;

[0017] The first type of prediction data is obtained by predicting the first type of data; the second type of prediction data can be obtained by predicting the second type of data or can be preset.

[0018] The city's ride-hailing saturation index was used as the initial data label.

[0019]

[0020] in, Let n be the ride-hailing saturation index for city n in year m. When, it indicates that the number of ride-hailing vehicles in city n is saturated; when When, it indicates that the number of ride-hailing vehicles in city n is not saturated; μ(m,n) is the average salary of full-time ride-hailing drivers in city n in year m; h(m,n) is the average working hours of full-time ride-hailing drivers in city n in year m; a(m,n) is the average salary of city n in year m.

[0021] The urban ride-hailing saturation judgment model includes a weighted K-fold cross-validation method; the weighted K-fold cross-validation method is used for model training in the data prediction layer.

[0022] The weighted K-fold cross-validation method includes a weighted error root that considers the time factor.

[0023]

[0024]

[0025] Where i represents different years; n is the total number of years in the data; δ i y is the time weighting factor; r is the root of the weighting error; i1 This is real city-level data, type 1; y i2 This is the first type of data for predicting cities.

[0026] The urban ride-hailing saturation judgment model includes the fluctuation cuckoo optimization algorithm:

[0027] w i t+1 =w i t +α i *s*(w i t -w Best t );

[0028]

[0029] Among them, w i t w represents the weight of the i-th parameter in the t-th iteration; Best t The parameter value represents the best value in the t-th iteration; s is the step size, which follows a Lévy distribution; α i F is the fluctuation step size factor of the i-th parameter; l is the adjustment parameter; F Best F represents the globally optimal nest fitness value. αThe value represents the fitness of an individual bird's nest. When the fitness of an individual bird's nest is close to the global optimal fitness, the fluctuation step size factor decreases; when the fitness of an individual bird's nest is far from the global optimal fitness, the fluctuation step size factor increases.

[0030] The data preprocessing includes a normalization operation using the max-min normalization method:

[0031]

[0032] Where, x i The initial data for normalization; max(x) is the maximum value in the same data dimension; min(x) is the minimum value in the same data dimension; x j This is the output value after normalization.

[0033] The urban ride-hailing saturation judgment model includes the Sig-ReLU activation function:

[0034]

[0035] Where x is the input data feature; e is the natural constant; and f is the processing result obtained after processing by the activation function.

[0036] A ride-hailing monitoring system includes a data collection module, a data prediction module, a data training module, and a result output module;

[0037] The data collection module is used to collect urban data and preprocess it to obtain the initial data of the city.

[0038] The data prediction module is used to perform regression prediction using the initial data to obtain a prediction dataset;

[0039] The data training module is used to train the prediction dataset to obtain the judgment result;

[0040] The result output module is used to output the judgment result.

[0041] The data collection module includes the city's ride-hailing saturation index;

[0042]

[0043] in, Let n be the ride-hailing saturation index for city n in year m. When, it indicates that the number of ride-hailing vehicles in city n is saturated; when When, it indicates that the number of ride-hailing vehicles in city n is not saturated; μ(m,n) is the average salary of full-time ride-hailing drivers in city n in year m; h(m,n) is the average working hours of full-time ride-hailing drivers in city n in year m; a(m,n) is the average salary of city n in year m.

[0044] The data prediction module includes an RVM prediction model and an LSTM prediction model for predicting the initial dataset.

[0045] The RVM prediction model is used to predict the initial dataset to obtain the RVM prediction dataset;

[0046] The LSTM prediction model is used to predict the initial dataset to obtain the LSTM prediction dataset;

[0047] The prediction dataset is obtained by weighting the RVM prediction dataset and the LSTM prediction dataset.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] 1. This invention proposes a city ride-hailing saturation judgment model to determine whether the number of ride-hailing vehicles in a city is saturated. The model comprises a data input layer, a data prediction layer, a data training layer, and a judgment output layer. The data prediction layer includes an RVM prediction model and an LSTM prediction model. These two prediction models are used to predict the data and perform a weighted sum to obtain a prediction dataset. The data training layer trains the input prediction dataset and judges whether the number of ride-hailing vehicles in the city is saturated based on the training results. This invention uses a parallel weighted summation of the RVM and LSTM prediction models, which can effectively predict based on historical city data, obtaining more accurate city prediction data, ensuring the accuracy of the city ride-hailing saturation judgment model, and obtaining more reliable ride-hailing saturation judgment results.

[0050] 2. This invention constructs a city ride-hailing saturation index to measure whether a city's ride-hailing market is saturated. The city ride-hailing saturation index includes three influencing factors: the ratio of the average wage of full-time ride-hailing drivers to the local city's average wage; the ratio of the average working hours of full-time ride-hailing drivers to the average working hours of the previous year; and the ratio of the average wage of full-time ride-hailing drivers to the average wage of the previous year. When the city ride-hailing saturation index is 0, it indicates that the city's ride-hailing fleet is saturated; when the city ride-hailing saturation index is 1, it indicates that the city's ride-hailing fleet is not saturated. This invention constructs a city ride-hailing saturation index from multiple influencing factors and uses the city ride-hailing saturation index as a data label to facilitate subsequent model training and adjustment, improving the accuracy of judging the saturation of the city's ride-hailing fleet.

[0051] 3. To address the parameter bias in the RVM and LSTM prediction models of the data prediction unit during training due to insufficient initial city data, this invention employs weighted K-fold cross-validation to process the initial data and training process, increasing the utilization of data samples. Furthermore, based on the obvious time-series characteristics of the initial city data used in this invention, and the relatively high correlation and influence between neighboring years, this invention improves the root mean square error in the classic K-fold cross-validation method by introducing a time weighting factor δ. i The time weighting factor can be adjusted and trained according to the differences between years to increase the influence of neighboring years. This invention, by employing weighted K-fold cross-validation, can effectively address training errors caused by insufficient data and leverage the temporal characteristics of the initial dataset to improve the prediction accuracy of the data prediction unit, thereby enhancing the overall performance of the urban network vehicle saturation model. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0053] Figure 2 This is a schematic diagram of the urban ride-hailing saturation judgment model of the present invention;

[0054] Figure 3 This is a schematic diagram of the training process of the urban ride-hailing saturation judgment model of the present invention;

[0055] Figure 4 This is a comparative experimental result diagram of the weighted k-fold cross-validation method of the present invention;

[0056] Figure 5 The diagram shows the experimental results of the optimization algorithm of this invention.

[0057] Figure 6 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0058] The proposed method for determining urban ride-hailing saturation will be further explained below according to the accompanying drawings provided in this invention. The technical explanations and experimental data involved in the specific implementation section are merely a specific embodiment of this technical solution. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0059] Ride-hailing services refer to online booking of taxis. These services utilize internet technology to build a platform that integrates supply and demand, breaking down information barriers between vehicles, drivers, and customers, thus supplementing the traditional taxi model. In its early stages, ride-hailing attracted a large number of people, either part-time or full-time, due to its flexible working hours, relatively high monthly income, and low entry barriers. However, with the influx of drivers, the market has gradually become saturated. The continuously increasing capacity has exceeded actual market demand, leading to problems such as low-price competition, market share grabbing, declining service quality, forced carpooling, and illegal operations.

[0060] Existing research on ride-hailing vehicle numbers primarily focuses on short-term demand forecasting, dividing cities into grid areas and predicting ride-hailing demand based on the number of orders in each grid area. However, this approach falls short when assessing ride-hailing vehicle saturation at the city level. Therefore, this invention provides a ride-hailing regulatory method and system that assesses and predicts ride-hailing vehicle saturation levels over a future period from a city-wide perspective.

[0061] To better understand whether the demand for ride-hailing services in a city is saturated, and to help ride-hailing platforms make strategic arrangements, this invention provides a ride-hailing supervision method and system. The method involves collecting ride-hailing-related data from cities across the country in recent years, training a city ride-hailing saturation judgment model, and obtaining the relevant parameters of the model. Then, the city ride-hailing saturation judgment model is used to identify predicted data for a city to determine whether the number of ride-hailing vehicles in the city will tend towards saturation in the future.

[0062] Please see Figures 1 to 6 This invention provides a method and system for regulating ride-hailing services, the technical solution of which is as follows:

[0063] A method for regulating ride-hailing services, characterized by comprising:

[0064] Collect city-level data and perform data preprocessing to obtain the initial dataset.

[0065] The initial dataset contains two types of data: a first type and a second type. The first type of data includes the city's permanent resident population, per capita GDP, urban traffic mileage, per capita car ownership, and urban weather index. The second type of data includes the number of ride-hailing vehicles in the city. All data in the initial dataset is for a specific city over one year.

[0066] Urban Resident Population: Residents refer to those who regularly reside in a region for more than six months. This mainly includes three types: 1) those residing in the region with local household registration or whose household registration is pending; 2) those residing in the region but having left their registered household location for more than six months; and 3) those with local household registration but having been away for less than six months or working / studying abroad. Urban GDP per Capita: This is the city's total GDP divided by the city's resident population. Urban Transportation Mileage: Urban transportation mileage refers to the sum of the lengths of all operating routes; including surface, underground, and elevated routes, but excluding non-operating routes such as turnaround lines, test lines, and connecting lines. Urban Car Ownership per Capita: This is the total number of cars registered in a city divided by the city's resident population. Urban Weather Index: This is the percentage of days with severe weather in a city throughout the year. Number of Ride-Hailing Vehicles in the City: This is the number of ride-hailing vehicles that operate for more than 210 days a year and have an average daily online time of more than 8 hours.

[0067] The data preprocessing includes a normalization operation using the max-min normalization method:

[0068]

[0069] Where, x i The initial data for normalization; max(x) is the maximum value in the same data dimension; min(x) is the minimum value in the same data dimension; x j This is the output value after normalization.

[0070] The initial dataset contains two types of labels: a city ride-hailing saturation index; when the city ride-hailing saturation index is 0, it indicates that the number of ride-hailing vehicles in the city is saturated; when the city ride-hailing saturation index is 1, it indicates that the number of ride-hailing vehicles in the city is not yet saturated.

[0071] The formula for the urban ride-hailing saturation index is as follows:

[0072]

[0073] in, Let n be the ride-hailing saturation index for city n in year m. When, it indicates that the number of ride-hailing vehicles in city n is saturated; when When the number of ride-hailing vehicles in city n is not saturated, μ(m,n) represents the average salary of full-time ride-hailing drivers in city n in year m; h(m,n) represents the average working hours of full-time ride-hailing drivers in city n in year m; and a(m,n) represents the average salary of city n in year m. Full-time ride-hailing drivers are defined as those who work more than 210 days a year and whose average daily platform online time exceeds 8 hours.

[0074] When judging and measuring whether the number of ride-hailing vehicles in a city is saturated, this invention mainly considers three aspects: comparing the average wage of full-time ride-hailing drivers with the local city average wage; comparing the average working hours of full-time ride-hailing drivers with the average working hours of the previous year; and comparing the average wage of full-time ride-hailing drivers with the average wage of the previous year. The product of these three ratios is then compared with 1. When the product is less than 1, the city's ride-hailing saturation index is 0, indicating that the number of ride-hailing vehicles in the city is saturated. When the product is greater than 1, the city's ride-hailing saturation index is 1, indicating that the number of ride-hailing vehicles in the city is not saturated.

[0075] This invention proposes a model for determining saturation in urban ride-hailing services. The working process is as follows: Figure 1 As shown.

[0076] Historical data at the city level was collected and preprocessed to obtain an initial dataset;

[0077] The initial dataset is input into the urban ride-hailing saturation judgment model; the initial dataset passes sequentially through the data input layer, data prediction layer, data training layer, and judgment output layer of the urban ride-hailing saturation judgment model, and outputs a judgment result; the judgment result is used to indicate whether the number of ride-hailing vehicles in the city is saturated; the structure of the urban ride-hailing saturation judgment model is as follows: Figure 2 As shown.

[0078] The data input layer is used to input the initial dataset into the urban ride-hailing saturation judgment model;

[0079] The data prediction layer is trained and predicted based on the initial dataset to obtain a prediction dataset; the data prediction layer includes an RVM prediction model and an LSTM prediction model; the RVM prediction model is used to predict the initial dataset to obtain an RVM prediction dataset; the LSTM prediction model is used to predict the initial dataset to obtain an LSTM prediction dataset; the RVM prediction dataset and the LSTM prediction dataset are weighted to obtain the prediction dataset.

[0080] The prediction dataset contains two types of prediction data: the first type and the second type. The first type of prediction data is derived from the first type of data. The second type of prediction data can be predicted from the second type of data or pre-defined. The predicted number of ride-hailing vehicles in a city can be predicted using a model based on historical data or manually set.

[0081] The data training layer includes a BP neural network; the BP neural network trains the prediction dataset and outputs training results; the BP neural network includes an improved cuckoo optimization algorithm for model optimization.

[0082] The judgment output layer makes a judgment based on the training results and outputs the judgment result.

[0083] The training process of the urban ride-hailing saturation judgment model is as follows: Figure 3 As shown.

[0084] The training of the model is divided into two parts: the RVM prediction model and the LSTM prediction model for data prediction, and the BP neural network for data judgment.

[0085] The model training method for the data prediction layer is as follows:

[0086] Historical data of the city is obtained, and the RVM and LSTM prediction models are trained using historical data from different years to obtain relevant parameters for predicting historical data of the city. RVM is a machine learning model based on a Bayesian framework, which obtains the relevant vector and weights by maximizing the marginal likelihood. LSTM is a special type of RNN network designed to solve long dependency problems and is suitable for time series prediction. Due to the existence of gradient problems, LSTM can only handle a certain order of magnitude of data; if the data is too large, gradient problems will occur, and the processing time will be too long. RVM, on the other hand, usually produces sparse models, making it faster on the test set while maintaining comparable generalization error. Therefore, this invention fits the above methods and weights the results of the two prediction methods.

[0087] Because the amount of data in the initial dataset is insufficient, the model training effect of the data prediction layer and the data training layer may be poor, and the parameters may be biased, affecting the final judgment accuracy of the urban network vehicle saturation judgment model. Therefore, this invention optimizes the data prediction layer and the data training layer respectively.

[0088] Therefore, this invention uses weighted k-fold cross-validation to process the samples. Weighted k-fold cross-validation can reuse data to improve the parameter accuracy of the prediction model.

[0089] The stability and fidelity of the evaluation results of cross-validation depend to a large extent on the value of k, which is affected by the training error. In the classic k-fold cross-validation method, the root mean square error is mainly used to measure the training error.

[0090]

[0091] Where i represents different years; n is the total number of years in the data; r0 is the root mean square error; y i1 This is real data; y i2 This is the data for prediction.

[0092] Since the city data used in this invention is time series data, correlation detection revealed that it has a high correlation with neighboring years. Therefore, this invention adjusts the root mean square error to the weighted root mean square error to increase the influence of neighboring years, thus obtaining the weighted k-fold cross-validation method.

[0093]

[0094]

[0095] Where i represents different years; n is the total number of years in the data; δ i y is the time weighting factor; r is the root of the weighting error; i1 This is real data; y i2 This is the data for prediction.

[0096] The steps of the k-fold cross-validation method are as follows:

[0097] First, the initial dataset R is randomly divided into k mutually exclusive subsets of the same size. Verification shows that K=10 yields the best results; thus, the data subsets R1, R2, R3…R are obtained. 10 ;

[0098] Nine samples are randomly selected as the training set each time, and the remaining sample is used as the test set to train the model. This process is repeated 10 times, with each validation dataset being different, resulting in parameter sets E1, E2, E3…E… 10 ;

[0099] For parameter sets E1, E2, E3…E 10 Calculate the mean to obtain the final model parameters.

[0100] To verify the actual effectiveness of the weighted K-fold cross-validation method, this invention uses the controlled variable method for research, including a first validation model using the weighted K-fold cross-validation method, a second validation model using the classic K-fold cross-validation method, and a third validation model without any processing method; all three models use the urban ride-hailing saturation judgment model, and only the training method in the models is different.

[0101] Using the initial dataset collected and established, after preprocessing, 80% of the data in the dataset is selected by random sampling for each experiment, and it is divided into an experimental training set and an experimental test set.

[0102] The first, second, and third validation models were trained using the experimental training set, and then tested using the experimental test set to obtain the accuracy rate of the urban ride-hailing saturation judgment model. The accuracy rate is the ratio of the number of times the urban ride-hailing saturation judgment model correctly judged the vehicle to the total number of judgments in the experimental test set data. The above experimental process was repeated 10 times, and the accuracy rate of each model was recorded.

[0103] The final experimental results are as follows Figure 4 As shown in the figure, the horizontal axis represents the number of experiments conducted, and the vertical axis represents the model's recognition accuracy.

[0104] Overall, the experimental results show that the first validation model using the weighted K-fold cross-validation method has the highest average accuracy of 94.05%, followed by the second validation model using the classic K-fold cross-validation method with an average accuracy of 91.71%, and finally the third validation model without any processing method, with an average accuracy of 86.78%.

[0105] The training steps of the BP neural network are as follows: First, obtain the historical data of the city and use the relationship between the average wage of ride-hailing drivers and the average wage of the city in each year as the data label; train the BP neural network with the historical data of the city and the corresponding labels to obtain the weight and threshold of the BP neural network; combine the two parts to obtain the urban ride-hailing saturation judgment model.

[0106] Similarly, due to the limited amount of data samples, BP neural networks may experience overfitting or underfitting, as well as local optima traps caused by unreasonable initial values. This invention addresses the parameter bias caused by insufficient data by improving the Cuckoo Optimization Algorithm. The Cuckoo Optimization Algorithm, among similar algorithms, features fewer parameters, faster convergence speed, and insensitivity to parameter changes, effectively mitigating local optima traps. Furthermore, to address the dilemma of balancing global search speed and accuracy caused by the fixed step size in the Cuckoo Optimization Algorithm, this invention proposes a fluctuating step size factor, enabling the Cuckoo Optimization Algorithm to achieve a balance between search speed and accuracy.

[0107] The Cuckoo Algorithm is an optimization problem-solving algorithm. The authors establish three assumptions: each cuckoo lays only one egg at a time, placing it in a randomly selected host nest; a greedy strategy is followed, with the best nest for each egg being retained generation after generation; and the number of available host nests is fixed, with a probability P (P∈(0,1)) that a cuckoo egg is found by a host. In this scenario, the host either discards the egg or abandons the nest altogether, building a completely new one. In summary, each egg in the nest represents a solution, and each cuckoo can only lay one egg. The algorithm uses a balanced combination of local random walks and global exploratory random walks, controlled by a switching parameter p.

[0108] The wave-cuckoo algorithm can be expressed by the following formula:

[0109] w i t+1 =w i t +α i *s*(w i t -w Best t );

[0110] w i t w represents the weight of the i-th bird's nest in the t-th iteration. Best t Let represent the best bird's nest in the t-th iteration, s be the step size, and follow a Lévy distribution. α i This is the fluctuation step size factor.

[0111]

[0112] In the formula, β is the scale attribute parameter that controls the random process, and its value ranges from 1 to 3. u and v are random numbers that follow a normal distribution.

[0113] In the standard Cuckoo algorithm, the step size is a fixed value, which is not conducive to the algorithm finding the optimal value. If the fixed step size is set to be large, it is beneficial for global search and improves the search speed, but it will reduce the search accuracy. If the fixed step size is set to be small, it is beneficial for local search and improves the search accuracy, but it will reduce the search speed and is prone to getting trapped in local optima.

[0114] This invention provides a fluctuation step size factor that can be adjusted based on data:

[0115]

[0116] α i F is the fluctuation step size factor for the i-th bird's nest; l is the adjustment parameter; F Best F represents the globally optimal nest fitness value. α For individual bird nest fitness values, and F Best <F α When the fitness of an individual bird's nest approaches the global optimal fitness, the fluctuation step size factor α... i Decrease the step size factor α to improve search accuracy; when the fitness of an individual nest is far from the global optimum, the step size factor α is adjusted. i Increase the size to improve search speed.

[0117] To verify the actual effectiveness of the improved cuckoo optimization algorithm, this invention employs the controlled variable method. The first optimization model using the improved cuckoo optimization algorithm is used as the experimental group, while the second optimization model using the cuckoo optimization algorithm, the third optimization model using the sparrow search algorithm, and the fourth optimization model using the classic BP neural network are used as control groups. All four optimization models use the urban ride-hailing saturation judgment model, and the only difference lies in the optimization algorithm.

[0118] Using the initial dataset that was collected and established, after data preprocessing, 80% of the data in the initial dataset was selected by random sampling for each experiment, and it was divided into an experimental training set and an experimental test set.

[0119] The first, second, third, and fourth optimized models were trained using an experimental training set, and then tested using an experimental test set. The accuracy rate of the urban ride-hailing saturation judgment model was obtained, whereby the accuracy rate is the ratio of the number of times the urban ride-hailing saturation judgment model correctly judged the data to the total number of judgments in the experimental test set. The above experimental process was repeated 10 times, and the accuracy rate of each model was recorded.

[0120] The final experimental results are as follows Figure 5 As shown in the figure, the horizontal axis represents the number of experiments conducted, and the vertical axis represents the model's recognition accuracy.

[0121] Overall, the experimental results show that the first optimization model using the improved cuckoo optimization algorithm has the highest average accuracy of 92.79%, followed by the third optimization model using the sparrow search algorithm with an average accuracy of 88.73%. Then comes the second optimization model using the cuckoo optimization algorithm with an average accuracy of 87.63%, and finally the fourth optimization model using the classic BP neural network with an average accuracy of 86.83%. The first optimization model using the improved cuckoo optimization algorithm has the best experimental performance.

[0122] A ride-hailing monitoring system includes a data collection module, a data prediction module, a data training module, and a result output module, such as... Figure 6 As shown.

[0123] The data collection module is used to collect urban data and preprocess it to obtain the initial data of the city.

[0124] The data prediction module is used to perform regression prediction using initial data to obtain a prediction dataset; the data prediction module includes an RVM prediction model and an LSTM prediction model; the RVM prediction model is used to predict the initial dataset to obtain an RVM prediction dataset; the LSTM prediction model is used to predict the initial dataset to obtain an LSTM prediction dataset; the RVM prediction dataset and the LSTM prediction dataset are weighted to obtain the prediction dataset;

[0125] The data training module is used to train the prediction dataset to obtain the judgment result; the data training module includes a BP neural network; the BP neural network is used to train the prediction dataset and output the training result; the BP neural network includes an improved cuckoo optimization algorithm for model optimization;

[0126] The result output module is used to output the judgment result.

[0127] Use the city's ride-hailing saturation index as a data label;

[0128]

[0129] in, Let n be the ride-hailing saturation index for city n in year m. When, it indicates that the number of ride-hailing vehicles in city n is saturated; when When, it indicates that the number of ride-hailing vehicles in city n is not saturated; μ(m,n) is the average salary of full-time ride-hailing drivers in city n in year m; h(m,n) is the average working hours of full-time ride-hailing drivers in city n in year m; a(m,n) is the average salary of city n in year m.

[0130] The above is a specific embodiment based on the technical solution of the present invention. Various changes made to the embodiment without departing from the basic principles and ideas of the present invention are within the protection scope of the present invention.

Claims

1. A method for regulating ride-hailing services, characterized in that, This includes a city ride-hailing saturation judgment model that determines the saturation of the number of ride-hailing vehicles in a city. The judgment process is as follows: Historical data of the city was collected and preprocessed to obtain an initial dataset. The initial dataset is input into the urban ride-hailing saturation judgment model, and the judgment result is output. The initial dataset passes through the data input layer, data prediction layer, data training layer, and judgment output layer of the urban ride-hailing saturation judgment model in sequence. The judgment result is used to indicate whether the number of ride-hailing vehicles in the city is saturated. The data input layer is used to input the initial dataset into the urban ride-hailing saturation judgment model; The data prediction layer is used to train and predict the initial dataset to obtain a predicted dataset; the data prediction layer includes an RVM prediction model and an LSTM prediction model; the RVM prediction model is used to predict the initial dataset to obtain an RVM predicted dataset; The LSTM prediction model is used to predict the initial dataset to obtain the LSTM prediction dataset; the RVM prediction dataset and the LSTM prediction dataset are weighted to obtain the prediction dataset; the data prediction layer includes weighted k-fold cross-validation. The data training layer is used to train the prediction dataset and output the training results; when the training result is 0, it indicates that the number of ride-hailing vehicles in the city is saturated; when the training result is 1, it indicates that the number of ride-hailing vehicles in the city is not saturated. The judgment output layer is used to output the training results.

2. The ride-hailing supervision method according to claim 1, characterized in that: The initial dataset includes a first type of data and a second type of data; The first category of data includes the city's permanent resident population, the city's GDP per capita, the city's traffic mileage, the city's car ownership per capita, and the city's weather index; The second category of data includes the number of ride-hailing vehicles in the city.

3. The method for regulating ride-hailing services according to claim 1, characterized in that: The prediction dataset contains a first type of prediction data and a second type of prediction data; The first type of prediction data is obtained based on the first type of data; The second type of predicted data can be obtained either by predicting the second type of data or by pre-setting it.

4. The method for regulating ride-hailing services according to claim 1, characterized in that: The urban ride-hailing saturation index is used as a data label for training the urban ride-hailing saturation judgment model. The city's ride-hailing saturation index includes three influencing factors: the ratio of the average wage of full-time ride-hailing drivers to the local city's average wage, the ratio of the average working hours of full-time ride-hailing drivers to the average working hours of last year, and the ratio of the average wage of full-time ride-hailing drivers to the average wage of last year. The formula for the urban ride-hailing saturation index is as follows: in, Let n be the ride-hailing saturation index for city n in year m. When, it indicates that the number of ride-hailing vehicles in city n is saturated; when When, it indicates that the number of ride-hailing vehicles in city n is not saturated; μ(m,n) is the average salary of full-time ride-hailing drivers in city n in year m; h(m,n) is the average working hours of full-time ride-hailing drivers in city n in year m; a(m,n) is the average salary of city n in year m.

5. The ride-hailing supervision method according to claim 3, characterized in that: The urban ride-hailing saturation judgment model includes a weighted K-fold cross-validation method; the weighted K-fold cross-validation method is used for model training of the data prediction layer under small sample conditions; The weighted K-fold cross-validation method includes a weighted error root that incorporates a time factor: Where i represents different years; n is the total number of years in the data; δ i y is the time weighting factor; r is the root of the weighting error; i1 This is real city-level data, type 1; y i2 This is the first type of data for predicting cities.

6. The method for regulating ride-hailing services according to claim 1, characterized in that: The urban ride-hailing saturation judgment model includes the wave cuckoo algorithm, which optimizes the BP neural network. The wave-cuckoo algorithm includes a wave step size factor: w i t+1 =w i t +α i *s*(w i t -w Best t ); Among them, w i t w represents the weight of the i-th parameter in the t-th iteration; Best t The parameter value represents the best value in the t-th iteration; s is the step size, which follows a Lévy distribution; α i F is the fluctuation step size factor of the i-th parameter; l is the adjustment parameter; F Best F represents the globally optimal nest fitness value. α For individual bird nest fitness values, and F Best <F α When the fitness of an individual bird's nest approaches the global optimal fitness, the fluctuation step size factor α... i Decrease; when the fitness of an individual bird's nest deviates from the global optimal fitness, the fluctuation step size factor α decreases. i Increase.

7. The method for regulating ride-hailing services according to claim 1, characterized in that: The urban ride-hailing saturation judgment model includes the Sig-ReLU activation function: Where x is the input data feature; e is the natural constant; and f is the processing result obtained after processing by the activation function.

8. A ride-hailing monitoring system, characterized in that: A ride-hailing monitoring system includes a data collection module, a data prediction module, a data training module, and a result output module; The data collection module is used to collect urban data and preprocess it to obtain the initial dataset of the city. The data prediction module is used to perform regression prediction using the initial dataset to obtain the prediction dataset; The data training module is used to train the prediction dataset to obtain the judgment result; The result output module is used to output the judgment result.

9. A ride-hailing monitoring system according to claim 8, characterized in that: The data collection module includes a city ride-hailing saturation index, which is used as a label for the initial data of the city. in, Let n be the ride-hailing saturation index for city n in year m. When, it indicates that the number of ride-hailing vehicles in city n is saturated; when When, it indicates that the number of ride-hailing vehicles in city n is not saturated; μ(m,n) is the average salary of full-time ride-hailing drivers in city n in year m; h(m,n) is the average working hours of full-time ride-hailing drivers in city n in year m; a(m,n) is the average salary of city n in year m.

10. A ride-hailing monitoring system according to claim 8, characterized in that: The data prediction module includes an RVM prediction model and an LSTM prediction model for predicting the initial dataset. The RVM prediction model is used to predict the initial dataset to obtain the RVM prediction dataset; The LSTM prediction model is used to predict the initial dataset to obtain the LSTM prediction dataset; The prediction dataset is obtained by weighting the RVM prediction dataset and the LSTM prediction dataset.

Images