A station flow prediction system for dockless shared electric motorcycles based on spatial association

Abstract

The present invention discloses a station flow prediction system for free-loading shared electric motorcycles based on spatial association, including a data acquisition and preprocessing module, the data acquisition and preprocessing module is connected to a station flow influencing factor analysis module, the station flow influencing factor analysis module is connected to a station flow prediction model construction module, the station flow prediction model construction module is connected to a model evaluation and optimization module, the model evaluation and optimization module is connected to a result output and application module, and the model evaluation and optimization module includes a point prediction model evaluation module. The present invention focuses on the spatial correlation between sites and introduces it into the point prediction and interval prediction models to verify whether the spatial correlation factors between sites have an improvement effect on the site flow prediction. The evaluation criteria of this system are more comprehensive, which further ensures the accuracy of the results.

Description

Technical Field

[0001] The present invention relates to the technical field of shared electric motorcycles, and specifically to a station flow prediction system for shared electric motorcycles without piles based on spatial association. Background Art

[0002] At present, the problems of urban traffic congestion and high carbon emissions have not been solved, which has reduced the quality of life of urban residents to a certain extent and also damaged the air quality of the city. In urban life, the promotion of shared travel can effectively alleviate the problems of urban traffic congestion and high carbon emissions. Due to its advantages in short-distance travel, shared travel has become an important connection tool between different modes of transportation such as buses and subways. The shared travel system mainly includes shared bicycles without battery systems and shared electric bicycles with battery systems. Shared bicycles include urban public bicycles with piles and shared bicycles without piles. Shared electric bicycles mainly exist in the form of piles without piles.

[0003] Unlike shared bicycles, the riding of dockless shared electric motorcycles with batteries is relatively labor-saving and more convenient. While providing urban residents with diversified travel methods, it also makes urban transportation more convenient and orderly. The only drawback is that dockless shared electric motorcycles have problems with battery replacement and vehicle scheduling. In order to better carry out this work, it is necessary to restrict car users to borrow and return vehicles at designated stations to prevent vehicles from being too scattered, increasing the cost of battery replacement and vehicle scheduling. However, there is a traffic tidal phenomenon in the travel of urban residents, which makes the flow of dockless shared electric motorcycles relatively unbalanced in different periods. The imbalance between supply and demand of station vehicles has caused some users to face problems such as "difficulty in borrowing and returning vehicles". Therefore, vehicle dispatchers are required to dispatch vehicles in advance so that the vehicles can be fully utilized. In order to achieve scientific and reasonable vehicle scheduling, the first prerequisite is to predict the traffic flow of the station. Whether the operator's scheduling decision is reliable and practical depends on the accuracy of the station traffic prediction. The station traffic prediction of dockless shared electric motorcycles is an important task in operation. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a station traffic prediction system for dockless shared electric motorcycles based on spatial correlation, which solves the problem of traffic tidal phenomenon in urban residents' travel, which makes the flow of dockless shared electric motorcycles relatively unbalanced in different periods and the imbalance between supply and demand of vehicles at stations.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a station flow prediction system for a dockless shared electric motorcycle based on spatial association, including a data acquisition and preprocessing module, the data acquisition and preprocessing module is connected to a station flow influencing factor analysis module, the station flow influencing factor analysis module is connected to a station flow prediction model construction module, the station flow prediction model construction module is connected to a model evaluation and optimization module, the model evaluation and optimization module is connected to a result output and application module, the model evaluation and optimization module includes a point prediction model evaluation module, the point prediction model evaluation module is for random forest and LightGB The point prediction model of M is connected, the point prediction model evaluation submodule is connected to the interval prediction model evaluation module, the interval prediction model evaluation module is connected to the model optimization unit, the model optimization unit adjusts the model parameters according to the model evaluation result, the result output and application module includes a prediction result output module, the prediction result output module organizes and outputs the prediction results of each prediction model, and displays the predicted value, true value and prediction interval of the site traffic in a visual manner, the prediction result output module is connected to the application decision support module, and the application decision support module provides decision support for the dispatching personnel of the dockless shared electric motorcycle according to the prediction result.

[0006] Preferably, the data acquisition and preprocessing module includes a data acquisition module, which is responsible for acquiring data from multiple sources. The data acquisition module is connected to a data preprocessing module, which cleans and processes the collected data.

[0007] Preferably, the site traffic influencing factor analysis module includes a time factor analysis module, the time factor analysis module is connected to a meteorological factor analysis module, and the meteorological factor analysis module is connected to a spatial correlation factor analysis module.

[0008] Preferably, the time factor analysis module is connected to a monthly fluctuation analysis unit, which analyzes the site traffic data over a long time span, draws a time series graph, studies the fluctuation of site traffic in different months, and reveals the seasonal variation of traffic. The monthly fluctuation analysis unit is connected to a periodic fluctuation analysis unit, which draws the site traffic fluctuation curves on different days and at different times of a week.

[0009] Preferably, the meteorological factor analysis module is connected to a weather type impact analysis unit, and the weather type impact analysis unit draws a bar chart of site traffic distribution under different weather types. The weather type impact analysis unit is connected to a meteorological factor impact analysis unit, and the meteorological factor impact analysis unit quantitatively analyzes the degree of influence of each meteorological factor on the site traffic by drawing a scatter plot and a fitted straight line between the meteorological factor and the site traffic.

[0010] Preferably, the spatial correlation factor analysis module is connected to a site spatial clustering unit, which calculates the Euclidean distance using the site longitude and latitude, clusters the sites using the k-means algorithm, and tries different k values ​​to evaluate the effect. The site spatial clustering unit is connected to a site spatial correlation impact analysis unit, which calculates the flow probability between associated sites.

[0011] Preferably, the site traffic prediction model construction module includes a random forest prediction model construction module, the random forest prediction model construction module is connected to a LightGBM prediction model construction module, and the LightGBM prediction model construction module is connected to a neural network quantile regression prediction model construction module.

[0012] Preferably, the random forest prediction model construction module is connected to a random forest parameter tuning unit, which tunes key parameters in the random forest model. The random forest parameter tuning unit is connected to a random forest model training and prediction unit, which uses a data set that has been preprocessed and feature engineered, trains a model based on the tuned parameters, and performs site traffic prediction on a test set.

[0013] Preferably, the LightGBM prediction model construction module is connected to a core parameter tuning unit, which adjusts the core parameters of the LightGBM model. The core parameter tuning unit is connected to a LightGBM model training and prediction unit, which uses the same data set to train the model according to the optimal parameters and predict the test set site traffic, and compares it with the prediction results of the random forest model.

[0014] Preferably, the neural network quantile regression prediction model construction module is connected to a neural network model parameter tuning unit, the neural network model parameter tuning unit determines the model parameters, and evaluates the model effect under different parameters through the AIC criterion. The neural network parameter tuning unit is connected to a neural network model training and prediction unit, and the neural network model training and prediction unit uses the processed data set to train the model, and uses a calculation formula to perform interval prediction on two sites and output the results.

[0015] Beneficial effects: Compared with the prior art, the station flow prediction system for shared electric motorcycles without docking stations based on spatial association provided by the present invention has the following characteristics:

[0016] 1. The present invention focuses on the spatial correlation between sites and introduces it into the point prediction and interval prediction models to verify whether the spatial correlation factors between sites have an improvement effect on the site flow prediction. Compared with the existing shared travel site flow prediction research, which only considers the impact of time and meteorological factors on site flow, the evaluation criteria of this system are more comprehensive, which further ensures the accuracy of the results.

[0017] 2. This paper applies the emerging machine learning algorithm LightGBM to the point prediction study of station traffic for dockless shared electric motorcycles, and explores the applicability of the LightGBM algorithm in this study, as well as its advantages in prediction accuracy and training efficiency, by empirically comparing it with the random forest model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a system flow chart of the present invention;

[0019] Figure 2 This is a flow chart of the site traffic influencing factor analysis module of the present invention;

[0020] Figure 3 A flow chart of a module for building a site traffic prediction model of the present invention;

[0021] Figure 4 This is a flow chart of the model evaluation and optimization module of the present invention;

[0022] Figure 5 It is a flow chart of the result output and application module of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Example:

[0025] Please see attached Figure 1 - Attachment Figure 5 An embodiment of the present invention provides a station traffic prediction system for a dockless shared electric motorcycle based on spatial association, including a data acquisition and preprocessing module, to which is connected a station traffic influencing factor analysis module, to which is connected a station traffic prediction model construction module, to which is connected a model evaluation and optimization module, and to which is connected a result output and application module.

[0026] The data collection and preprocessing module includes a data collection module, which is responsible for acquiring data from multiple sources, including site information of shared electric motorcycles, detailed order data, and historical meteorological data, providing a rich and practical data foundation for subsequent analysis;

[0027] Site information includes: site ID, name, address, longitude and latitude; order details include order ID, vehicle number, lock / unlock time, start and end site ID; historical meteorological data includes weather type, temperature, humidity, wind speed, visibility, and rainfall;

[0028] The data acquisition module is connected to a data preprocessing module, which cleans and processes the collected data.

[0029] The site traffic influencing factor analysis module includes a time factor analysis module, the time factor analysis module is connected to a meteorological factor analysis module, and the meteorological factor analysis module is connected to a spatial correlation factor analysis module.

[0030] The time factor analysis module is connected to a monthly fluctuation analysis unit, which analyzes the site flow data over a long period of time, draws a time series graph, studies the fluctuation of site flow in different months, and reveals the seasonal variation of flow, including low flow in winter and high flow in summer.

[0031] The monthly fluctuation analysis unit is connected to the periodic fluctuation analysis unit, which draws the site traffic fluctuation curves at different days and times of the week, and analyzes the impact of weekdays, weekends, and different time periods on traffic, including obvious peaks in the morning and evening on weekdays and gentle peaks on weekends.

[0032] The meteorological factor analysis module is connected to a weather type impact analysis unit, which draws a bar graph of site traffic distribution under different weather types, analyzes the impact of weather conditions on site traffic, and finds a positive correlation between weather conditions and site traffic. Weather conditions include cloudy, overcast, foggy, blowing sand, sunny, heavy rain, light rain, light snow, and sleet.

[0033] The weather type impact analysis unit is connected to a meteorological factor impact analysis unit. The meteorological factor impact analysis unit draws a scatter plot of meteorological factors and station flow and a fitted straight line to quantitatively analyze the impact of each meteorological factor on the station flow. Meteorological factors include temperature, humidity, visibility, wind speed, cloud cover, and rainfall. The impact of meteorological factors on station flow includes positive correlation with temperature and negative correlation with humidity. A prediction model is constructed using a calculation formula to evaluate the prediction effect of the meteorological factor-related model. The calculation formula is:

[0034]

[0035] It is used to measure the difference between the predicted value and the true value of the meteorological factor-related prediction model. The smaller the value, the higher the prediction accuracy. It can help determine the accuracy of the model when evaluating the impact of meteorological factors on flow prediction.

[0036]

[0037] It directly reflects the average absolute deviation between the predicted value and the true value, and is used to evaluate the accuracy of the prediction model related to meteorological factors;

[0038]

[0039] It is used to measure the degree of fit of the meteorological factor-related model to the data. The value range is between 0 and 1. The closer it is to 1, the better the fit.

[0040] The spatial correlation factor analysis module is connected to a site spatial clustering unit, which uses the longitude and latitude of the site to calculate the Euclidean distance and uses the k-means algorithm to cluster the sites. Different k values ​​are tried to evaluate the effect, including k=3, 4, 5, and 6. Finally, it is determined that k=4 has a better effect. The site categories are divided to determine the associated sites.

[0041] The site space clustering unit is connected to the site space correlation impact analysis unit, which calculates the flow probability between the associated sites. The specific calculation formula is:

[0042]

[0043] The vehicle inflow from the associated station that affects the central station is obtained by weighted summation of the vehicle outflow (station flow) at the previous moment. The weight refers to the probability of the associated station flowing into the central station. The vehicle inflow from the associated station to the target station at all times is recorded as the spatial association sequence that affects the central station.

[0044] In order to calculate the spatial association sequence of site A, the following calculation method is given for the sequence:

[0045] (1) The current time is time t. Extract the vehicle outflow (station flow) of the associated station i at time t-1, which is recorded as where i∈[1,2,...,n].

[0046] (2) Calculate the probability of a vehicle flowing from the associated site to site A:

[0047] In order to further evaluate the flow correlation between similar sites, the flow probability between sites is calculated based on the proportion of the number of flows in the same period. The following formula measures the flow probability from site i (associated site) to site A (target site) at time t. Considering that the flow of vehicles takes about one cycle, the calculation rule is the number of all vehicles that started from site i and stopped at site A at time t-1 in history, and the outflow volume is divided by the number of all vehicles that flowed out of site i at time t-1 in history.

[0048]

[0049] in, represents the number of vehicles that departed from station i at time t-1 but eventually flowed into station A. Represents the number of vehicles that have left the station i in history at time t-1

[0050] (3) Weighted sum of traffic of associated sites:

[0051]

[0052] in, represents the number of all associated sites flowing into site A at time t, represents the probability of the associated site flowing into site A at time t-1, Represents the outflow of the associated site at time t-1.

[0053] At this point, according to the weighted summation formula of the traffic of the associated site at the previous moment, the associated sequence value of site A at each moment t can be calculated, and the sequence composed of these values ​​is the spatial association sequence of site A. In order to further explore the impact of the spatial association sequence of site A on the site traffic at the current moment, the correlation coefficient is used here to measure the correlation between the two.

[0054] At present, the correlation coefficients commonly used to measure the correlation between numerical variables are mainly Pearson correlation coefficients, and the correlation coefficients used to measure the correlation between ordinal variables are Spearman correlation coefficients. Both spatial association series and station flow series are numerical variables, so the Pearson correlation coefficient is selected to provide a basis for us to explore the correlation between numerical variables. The expression of the Pearson correlation coefficient is shown in the following formula:

[0055]

[0056] The value of the Pearson correlation coefficient is between -1 and 1. When the value is greater than 0, it indicates that the relationship between the variables is positively correlated. When it is less than 0, it indicates that the relationship between the variables is negatively correlated. The absolute value of the Pearson correlation coefficient reflects the strength of the correlation between the variables. Generally, when its absolute value is greater than or equal to 0.8, it indicates that the correlation between the variables is high. When the absolute value is greater than or equal to 0.5 and less than 0.8, it indicates that the correlation between the variables is medium. When the absolute value is greater than or equal to 0.3 and less than 0.5, it indicates that there is a weak correlation between the variables. When it is less than 0.3, it indicates that the correlation between the variables is extremely weak or non-existent.

[0057] After obtaining the spatial correlation sequence of site A through the above calculation method, the Pearson correlation coefficient of the spatial correlation sequence of site A to the flow sequence of site A at the current moment is further calculated. According to this method, the Pearson correlation coefficient of the spatial correlation sequence of each site and the flow sequence of the current site is calculated.

[0058] The site traffic prediction model construction module includes a random forest prediction model construction module, the random forest prediction model construction module is connected to the LightGBM prediction model construction module, and the LightGBM prediction model construction module is connected to the neural network quantile regression prediction model construction module.

[0059] The random forest prediction model construction module is connected to a random forest parameter tuning unit, which tunes the key parameters (n_est imators, max_features, max_depth, min_sample_split, min_sample_leaf) in the random forest model. Taking the RMSE value of the training as the standard, the optimal parameter value (n_est imators = 50, max_depth = 10) is determined through experiments to avoid underfitting or overfitting of the model and improve the prediction accuracy; the random forest parameter tuning unit is connected to a random forest model training and prediction unit, which uses the data set processed by preprocessing and feature engineering, trains the model according to the tuned parameters, and predicts the site traffic on the test set.

[0060] The LightGBM prediction model construction module is connected to a core parameter tuning unit, which adjusts the core parameters of the LightGBM model (num_length, num_iterations, feature_fraction, bagging_fraction, length_rate, max_depth, max_bin). The grid search parameter tuning method is used in combination with manual debugging to narrow the parameter value range, and the optimal parameter combination (num_length = 35, length_rate = 0.05) is selected through 5-fold cross validation to improve the model prediction accuracy and training efficiency. This is one of the innovations of this system. The emerging algorithm LightGBM is applied to point prediction research, and its applicability and advantages are explored by comparing with random forests. The core parameter tuning unit is connected to the LightGBM model training and prediction unit. The LightGBM model training and prediction unit uses the same data set to train the model according to the optimal parameters and predict the site traffic of the test set. The prediction results are compared with the random forest model to show the advantages of LightGBM in prediction accuracy.

[0061] The neural network quantile regression prediction model construction module is connected to a neural network model parameter tuning unit, which determines the model parameters (number of neural network layers, number of hidden layer nodes, maximum number of iterations, penalty parameters), evaluates the model effect under different parameters through the AIC criterion, selects representative quantile points (0.1, 0.5, 0.9) by the grid search method, and determines the optimal parameter values ​​(the number of hidden layer nodes is 5 and the penalty parameter is 0.9) to ensure the accuracy of the model. The neural network parameter tuning unit is connected to a neural network model training and prediction unit, which uses the processed data set to train the model, predicts the site traffic at different quantile points (0.05, 0.1, 0.5, 0.9, 0.95), constructs prediction intervals with different confidence levels (80%, 90%), and uses the calculation formula to perform interval prediction on the two sites and output the results. The specific calculation formula is:

[0062]

[0063] Where PICP represents the coverage probability of the prediction interval, I t is the indicator function, when the true value y t In the prediction interval Internal Time I t =1, otherwise I t =0, used to measure the proportion of the predicted interval containing the true value. The higher the proportion, the higher the reliability of the interval prediction;

[0064]

[0065] Where PI NAW represents the normalized width of the prediction interval, and They are the maximum and minimum values ​​of the true value in the sample, respectively, and are used to measure the width of the prediction interval. The smaller the value, the narrower the interval width and the more accurate the prediction. The differences in these indicators between the models that introduce and exclude spatial correlation factors are compared to determine the role of spatial correlation factors in improving interval prediction.

[0066] The model evaluation and optimization module includes a point prediction model evaluation module. The point prediction model evaluation module uses the root mean square error (RMSE), mean absolute error (MAE) and determination coefficient (R2) evaluation indicators to evaluate the model prediction effect for the point prediction models of random forest and LightGBM, accurately quantify the difference between the model prediction value and the true value and the degree of fit of the model to the data, and help determine which model or model configuration performs better in the point prediction of the flow of dockless shared electric motorcycle stations. The point prediction model evaluation submodule is connected to the interval prediction model evaluation module. For the interval prediction model of the neural network quantile regression, the interval prediction model evaluation module uses the coverage probability of the prediction interval (PI CP) and the normalized width of the prediction interval (PI The NAW) indicator is used to evaluate the interval prediction effect, and the RMSE and MAE are used to evaluate the median point prediction effect. Through the calculation and comparison of these formulas, we can deeply understand the value of spatial correlation factors in interval prediction, as well as the prediction performance of the model under different circumstances. The interval prediction model evaluation module is connected to a model optimization unit. The model optimization unit optimizes the model according to the model evaluation results by adjusting model parameters, increasing the amount of data, and improving feature engineering to improve the model performance and make it more accurately predict the flow of dockless shared electric motorcycle stations.

[0067] The result output and application module includes a prediction result output module. The prediction result output module organizes and outputs the prediction results of each prediction model (random forest, LightGBM point prediction model and neural network quantile regression interval prediction model), and displays the predicted value, true value and prediction interval (interval prediction model) of the site traffic in a visual way (line graph, chart, etc.), so that users can intuitively understand the prediction results. The prediction result output module is connected to an application decision support module. The application decision support module provides decision support for the dispatchers of the dockless shared electric motorcycle vehicles according to the prediction results, including adding vehicles to the dispatch when there are sufficient vehicles but the flow between sites is uneven, referring to the prediction interval with high coverage and wide width; when there are few vehicles but the dispatch is balanced, referring to the interval with low coverage and narrow width to help arrange vehicles reasonably, improve operational efficiency, and realize the actual application value of the system.

[0068] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A station flow prediction system for dockless shared electric motorcycles based on spatial association, characterized by: It includes a data acquisition and preprocessing module, the data acquisition and preprocessing module is connected to a site traffic influencing factor analysis module, the site traffic influencing factor analysis module is connected to a site traffic prediction model construction module, the site traffic prediction model construction module is connected to a model evaluation and optimization module, and the model evaluation and optimization module is connected to a result output and application module; the model evaluation and optimization module includes a point prediction model evaluation module, the point prediction model evaluation module is for the point prediction model of random forest and LightGBM, the point prediction model evaluation submodule is connected to an interval prediction model evaluation module, the interval prediction model evaluation module is connected to a model optimization unit, the model optimization unit adjusts the model parameters according to the model evaluation result to optimize the model, the result output and application module includes a prediction result output module, the prediction result output module organizes and outputs the prediction results of each prediction model, and displays the predicted value, true value and prediction interval of the site traffic in a visual manner; the prediction result output module is connected to an application decision support module, and the application decision support module provides decision support for the dispatching personnel of the dockless shared electric motorcycle vehicle according to the prediction result.

2. The station flow prediction system for dockless shared electric motorcycles based on spatial association according to claim 1 is characterized by: The data acquisition and preprocessing module includes a data acquisition module, which is responsible for acquiring data from multiple sources. The data acquisition module is connected to a data preprocessing module, which cleans and processes the collected data.

3. The station flow prediction system for dockless shared electric motorcycles based on spatial association according to claim 1 is characterized by: The site traffic influencing factor analysis module comprises a time factor analysis module, the time factor analysis module is connected to a meteorological factor analysis module, and the meteorological factor analysis module is connected to a space correlation factor analysis module.

4. The station flow prediction system for dockless shared electric motorcycles based on spatial association according to claim 3 is characterized by: The time factor analysis module is connected to a monthly fluctuation analysis unit, which analyzes the site traffic data over a long period of time, draws a time series graph, studies the fluctuation of site traffic in different months, and reveals the seasonal variation of traffic. The monthly fluctuation analysis unit is connected to a periodic fluctuation analysis unit, which draws site traffic fluctuation curves on different days and at different times of a week.

5. The station flow prediction system for dockless shared electric motorcycles based on spatial association according to claim 4 is characterized by: The meteorological factor analysis module is connected to a weather type impact analysis unit, which draws a bar chart of site traffic distribution under different weather types. The weather type impact analysis unit is connected to a meteorological factor impact analysis unit, which quantitatively analyzes the impact of each meteorological factor on the site traffic by drawing a scatter plot and a fitted straight line between the meteorological factor and the site traffic.

6. The station flow prediction system for dockless shared electric motorcycles based on spatial association according to claim 3 is characterized by: The spatial correlation factor analysis module is connected to a site spatial clustering unit, which calculates the Euclidean distance using the site longitude and latitude, clusters the sites using the k-means algorithm, and tries different k values ​​to evaluate the effect. The site spatial clustering unit is connected to a site spatial correlation impact analysis unit, which calculates the flow probability between associated sites.

7. The station flow prediction system for dockless shared electric motorcycles based on spatial association according to claim 1, characterized in that: The site traffic prediction model construction module includes a random forest prediction model construction module, the random forest prediction model construction module is connected to a LightGBM prediction model construction module, and the LightGBM prediction model construction module is connected to a neural network quantile regression prediction model construction module.

8. The station flow prediction system for dockless shared electric motorcycles based on spatial association according to claim 1, characterized in that: The random forest prediction model construction module is connected to a random forest parameter tuning unit, which tunes key parameters in the random forest model. The random forest parameter tuning unit is connected to a random forest model training and prediction unit, which uses a data set that has been preprocessed and feature engineered, trains a model based on the tuned parameters, and performs site traffic prediction on a test set.

9. The station flow prediction system for dockless shared electric motorcycles based on spatial association according to claim 1, characterized in that: The LightGBM prediction model construction module is connected to a core parameter tuning unit, which adjusts the core parameters of the LightGBM model. The core parameter tuning unit is connected to a LightGBM model training and prediction unit, which uses the same data set to train the model according to the optimal parameters and predict the test set site traffic, and compares it with the prediction results of the random forest model.

10. The station flow prediction system for dockless shared electric motorcycles based on spatial association according to claim 7, characterized in that: The neural network quantile regression prediction model construction module is connected to a neural network model parameter tuning unit, which determines model parameters and evaluates model effects under different parameters through the AIC criterion. The neural network parameter tuning unit is connected to a neural network model training and prediction unit, which uses the processed data set to train the model, and uses a calculation formula to perform interval prediction on two sites and output the results.

Images