Shared bicycle tide point identification and rental return demand prediction method

Abstract

Most of the existing shared bicycle tidal researches tend to be macroscopic by using the shared bicycle hotspot area identified by the spatio-temporal clustering method, and the setting of clustering parameters has a great influence on the results, which is difficult to meet the application requirements of refined scheduling. The existing shared bicycle demand prediction method does not take into account the capacity of a single electronic fence, the difference between rental and return demand, which affects the prediction accuracy. Therefore, the invention aims to propose a small-scale shared bicycle tidal point identification method based on shared bicycle order data and electronic fence spatial data, and further utilize the K nearest neighbor (KNN) algorithm to calculate the related information between adjacent electronic fences, and construct a KNN-LightGBM model based on the fusion of space-time characteristics to predict the rental and return demand of the tidal point, thereby improving the prediction accuracy. The simulation data set verifies the effectiveness of the method.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of spatial information technology and spatial big data analysis, and particularly relates to a shared bicycle tide point identification and rental demand prediction method. BACKGROUND

[0002] With the rise of Internet rental bicycles (shared bicycles), "shared bicycles + subways" and "shared bicycles + buses" have become important connection modes for urban commuting. Shared bicycles are rapidly gathered or rented between hot areas (schools, office parks, bus stops, subway stations, residential areas, etc.) during morning and evening rush hours, resulting in local supply and demand imbalance in shared bicycle parking lots, which is referred to as the tide phenomenon of shared bicycles. This phenomenon has become a "pain point" and "difficulty" for the management and efficient scheduling of shared bicycle resources. Therefore, identifying electronic fences (tide points) with tide phenomenon characteristics and predicting rental demand are of great significance for the orderly and standardized development of shared bicycles and the optimization of people's bicycle experience.

[0003] In the aspect of shared bicycle tide point identification, existing research mainly uses spatiotemporal clustering methods to explore shared bicycle rental hot areas [1-3] . In the aspect of shared bicycle rental demand prediction, existing research mainly falls into two categories: (1) regression prediction models based on historical data [4,5] , mainly based on existing shared bicycle operation system travel data, using statistical learning methods for prediction; (2) prediction models based on machine learning [6,7] , mainly considering the time series characteristics of travel demand and external influencing factors (weather, population, land use, building environment factors) [8,9] , using machine learning methods for modeling and predicting rental demand.

[0004] [1] Jiang X, Bai L B, Lou X Y, et al. Shared Bicycle Tide Feature Mining and Demand Prediction Based on Multi-scale Spatiotemporal Clustering[J]. Journal of Geomatics Science, 2022, 24(6): 1047-1060.

[0005] [2] Xu W. Shared Bicycle Hot Area Identification and Demand Prediction Based on Machine Learning[J]. Comprehensive Transportation, 2019, 41(5): 29-34.

[0006] [3] Hui Y, Xie Y K, Yu Q, et al. Hotspots identification and classification of dockless bicycle sharing service under electric fence circumstances[J]. Journal of Advanced Transportation, 2022, 2022:5218254.

[0007] [4] Lin Y P, Dou W F. Short-term prediction method of urban public bicycle demand based on ARIMA model[J]. Journal of Nanjing Normal University(Natural Science Edition), 2016, 16(3):36-40.

[0008] [5] Kaltenbrunner A, et al. Urban cycles and mobility patterns: Exploring and predicting trends in a bicycle-based public transport system[J]. Pervasive and Mobile Computing, 2010, 6(4):455-466.

[0009] [6] Karlaftis M G, et al. Statistical methods versus neural networks in transportation research: Differences, similarities and some insights[J]. Transportation Research Part C: Emerging Technologies, 2011, 19(3):387-399.

[0010] [7] E S V. Using data mining techniques for bike sharing demand prediction in metropolitan city[J]. Computer Communications, 2020, 153:353-366.

[0011] [8]Faghih-Imani A, et al. How land-use and urban form impact bicycle flows: Evidence from the bicycle-sharing system (BIXI) in Montreal [J]. Journal of Transport Geography, 2014, 41: 306-314.

[0012] [9]Vlahogianni E I, et al. Short-term traffic forecasting: Where we are and where we’re going [J]. Transportation Research Part C: Emerging Technologies, 2014, 43: 3-19.

[0013] The deficiencies of the above prior art mainly include: (1) the shared bicycle hot area identified by the spatio-temporal clustering method tends to be macroscopic, and the setting of the clustering parameter has a great influence on the result, which is difficult to meet the fine scheduling application demand; (2) the existing shared bicycle demand prediction method does not take into account the capacity of a single electronic fence, the difference between rental and return demand, which affects the prediction accuracy. SUMMARY

[0014] With the rise of Internet rental bicycles (shared bicycles), "shared bicycles + subways" and "shared bicycles + buses" have become the main connection modes for urban commuting, but the "tidal effect" of shared bicycles has become a "pain point" and "difficulty" for shared bicycle management and resource allocation. Therefore, identifying electronic fences with tidal phenomenon characteristics (tidal points) and predicting rental and return demand are of great significance for the orderly and standardized development of shared bicycles and the optimization of people's bicycle experience. However, most of the existing shared bicycle tidal research uses spatio-temporal clustering method to identify shared bicycle hot areas, which tends to be macroscopic, and the setting of the clustering parameter has a great influence on the result, which is difficult to meet the fine scheduling application demand; the existing shared bicycle demand prediction method does not take into account the capacity of a single electronic fence, the difference between rental and return demand, which affects the prediction accuracy. Therefore, the present application aims to propose a small-scale shared bicycle tidal point identification method based on shared bicycle order data and electronic fence spatial data, and further use K-nearest neighbor (KNN) algorithm to calculate the related information between adjacent electronic fences, and construct a KNN-LightGBM model based on the fusion of spatio-temporal features to predict the rental and return demand of tidal points, so as to improve the prediction accuracy. The simulation data set verifies the effectiveness of the method of the present application.

[0015] In view of the defects and deficiencies of the prior art, the present application aims to provide a small-scale shared bicycle tidal point identification method based on shared bicycle order data and electronic fence space data, and further utilize the K-nearest neighbor (KNN) algorithm to calculate the correlation information between adjacent electronic fences, construct a KNN-LightGBM model based on fused spatiotemporal features to predict the tidal point rental demand, thereby improving the prediction accuracy.

[0016] Based on shared bicycle order data and electronic fence space data, the present application proposes a spatiotemporal model for identifying shared bicycle tidal points, and considers the spatiotemporal features and spatial correlation of electronic fences, and proposes a KNN-LightGBM rental demand prediction model that fuses spatiotemporal features. The main technical implementation route includes:

[0017] (1) Collect and preprocess shared bicycle order and electronic fence data;

[0018] (2) Estimate the rental and return of electronic fence vehicles;

[0019] (3) Identify shared bicycle tidal points based on spatiotemporal model;

[0020] (4) Extract adjacent electronic fence correlation information;

[0021] (5) Predict the tidal point rental demand based on the KNN-LightGBM model that fuses spatiotemporal features.

[0022] The technical solution adopted by the present application to solve its technical problems is:

[0023] A shared bicycle tidal point identification and rental demand prediction method, characterized by the following steps:

[0024] Step S1: Collect and preprocess shared bicycle order and electronic fence data;

[0025] Step S2: Estimate the rental and return of electronic fence vehicles;

[0026] Step S3: Identify shared bicycle tidal points based on spatiotemporal model;

[0027] Step S4: Extract adjacent electronic fence correlation information;

[0028] Step S5: Predict the tidal point rental demand based on the KNN-LightGBM model that fuses spatiotemporal features.

[0029] Further, in step S1, the collected order data and electronic fence data are processed, specifically including the following steps:

[0030] Step S11: Remove outliers: remove orders with excessively large or small trip distance and duration;

[0031] Step S12: electronic fence center point and parking space number estimation: calculate the area and center point position coordinates of each electronic fence, and estimate the number of parking locations.

[0032] Further, step S2 specifically includes the following steps:

[0033] Step S21: extract the starting point and ending point location information of the shared bicycle order;

[0034] Step S22: match the order starting point and ending point to the electronic fence using the hierarchical navigable small world graph algorithm;

[0035] Step S23: calculate the rental and return demand of each electronic fence in each time period.

[0036] Further, step S3 specifically includes the following steps:

[0037] Step S31: determine the shared bicycle tidal point identification rule: determine the tidal point identification rule according to the basic principle of "a large number of shared bicycles being rented or returned in a short period of time, resulting in no bicycles available for rent or no parking spaces available for parking in the electronic fence";

[0038] Step S32: build the lack of car / parking index: build the lack of car / parking index according to the state of the electronic fence in a certain time period;

[0039] Step S33: identify tidal points: identify shared bicycle tidal points according to the lack of car / parking index.

[0040] Further, step S4 specifically includes the following steps:

[0041] Step S41: set the correlation weight between electronic fences: set the correlation weight between electronic fences according to the spatial distance and threshold value between electronic fences;

[0042] Step S42: build a vehicle rental and return time series matrix: build a vehicle rental and return time series matrix according to the historical rental and return order data of the electronic fence;

[0043] Step S43: calculate the correlation of adjacent electronic fences: use KNN algorithm to calculate the correlation degree between the target electronic fence and the adjacent electronic fence in the local range in terms of historical rental and return demand.

[0044] Further, step S5 specifically includes the following steps:

[0045] Step S51: extract key features affecting shared bicycle rental and return demand: build a spatio-temporal feature matrix and extract key features using principal component analysis PCA;

[0046] Step S52: constructing model input features: fusing the key features extracted by PCA and the correlation degree of adjacent electronic fences calculated by KNN as input features;

[0047] Step S53: predicting tide point rental and return demand: based on the input features, using the LightGBM algorithm to predict the tide point rental and return demand.

[0048] Further, in step S33, identifying the shared bicycle tide point based on the lack of car / lack of parking index specifically includes:

[0049] Step S331: determining the shared bicycle tide point identification rule

[0050] Defining the shared bicycle tide point as: an electronic fence in which there is no car to rent or no parking space to park due to a large number of shared bicycle rentals or returns within a time T;

[0051] Step S32: dividing the electronic fence state type

[0052] ① No shared bicycles: a(i,j,k) = 0

[0053] ② Only a small number of shared bicycles: 0 < a(i,j,k) ≤ f × p(i)

[0054] ③ There are sufficient shared bicycles and parking spaces: p(i) × f < a(i,j,k) ≤ p(i) × g

[0055] ④ Only a small number of parking spaces: p(i) × g < a(i,j,k) ≤ p(i)

[0056] ⑤ Mild accumulation: 0 < b(i,j,k) ≤ p(i) × h

[0057] ⑥ Severe accumulation: p(i) × h < b(i,j,k)

[0058] Wherein: a(i,j,k) is the number of shared bicycles available for rent in the i-th electronic fence on the j-th day in the k-th time period; b(i,j,k) is the number of parking spaces lacking in the i-th electronic fence on the j-th day in the k-th time period; p(i) is the total number of parking spaces in the i-th electronic fence; T, f, g, h are variable parameters;

[0059] Step S333: constructing the lack of car / lack of parking index

[0060] The lack of car index c(i,j,k) and the lack of parking index d(i,j,k) of the i-th electronic fence on the j-th day in the k-th time period are defined as shown in equations (1) and (2):

[0061]

[0062]

[0063] Step S334: Calculate the missing bike / parking index of the electronic fence

[0064] Based on the definition of step S333, the missing bike / parking index of the electronic fence is calculated as shown in equations (3), (4), (5) and (6):

[0065]

[0066]

[0067]

[0068]

[0069] C(i,j,k) represents the total missing bike index of the i-th electronic fence, D(i,j,k) represents the total missing parking index of the i-th electronic fence; E(i,j,k) represents the missing bike index of the shared bicycle system at the j-th day and the k-th time period, and F(i,j,k) represents the missing parking index of the shared bicycle system at the j-th day and the k-th time period;

[0070] Step S335: Identify the tidal point

[0071] According to the calculated missing bike / parking index, the shared bicycle tidal point is identified.

[0072] Further, in step S4,

[0073] Step S41 is specifically:

[0074] Considering the spatial distance between the electronic fences, the correlation weight between the electronic fences is represented by D(x a , x b ), as shown in equation (7):

[0075]

[0076] dist(x a ,x b ) represents the distance between the electronic fences x a and x b ; R is the distance threshold value; when the distance between two electronic fences is less than or equal to the size of the threshold value R, the correlation weight coefficient of the two electronic fences is 1, otherwise it is 0;

[0077] Step S42 is specifically:

[0078] A time series matrix X is established through the historical rental data of the electronic fence, as shown in equation (8), x a t represents the electronic fence x aThe number of rented bikes is counted at t, and the return demand time series matrix is constructed according to the same rule:

[0079]

[0080] Step S43 is specifically:

[0081] Electronic fence x a And x b The similarity score Sim(x a , x b ) is calculated by formula (9) to obtain the electronic fence most similar to the tidal point; finally, the evaluation index of the prediction model: root mean square error RMSE, mean absolute error MAE, mean coefficient of determination R 2 And correlation coefficient r, as shown in formulas (10), (11), (12) and (13), determine the optimal value of parameter K:

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] Further, in step S5,

[0088] Step S51 is specifically:

[0089] The spatio-temporal data including at least: shared bicycle orders, city information points, roads, population, land use, air temperature, wind speed, the influence on rental and return demand is comprehensively considered, the spatio-temporal characteristics of the tidal point are analyzed, and the PCA algorithm of SPSS software is used to extract z main components;

[0090] Step S53 specifically includes:

[0091] Step S531: Construct a training set;

[0092] The training set is constructed by using the shared bicycle rental and return history data, and the features obtained by construction are input into the LightGBM algorithm;

[0093] Step S532: Train and optimize the LightGBM model:

[0094] First, linearly combine M weak regression trees into a strong regression tree, as shown in formula (14):

[0095]

[0096] Where F(x) is the sum of the final output values, f m (x) is the output value of the m-th weak regression tree;

[0097] Then, the input features are merged using the histogram algorithm: the continuous feature values ​​are divided into K integers and a histogram of width K is constructed. When traversing the data, the discretized values ​​are used as indices and accumulated in the histogram to find the best decision tree split point.

[0098] Finally, a leaf-wise growth strategy with depth restrictions is used to find the leaf with the maximum splitting gain among all the current leaves and split it, and a maximum depth restriction is set.

[0099] Step S533: Predict and output the rental and return demand at tidal points.

[0100] The key design features of this invention include:

[0101] 1. A bike shortage / parking shortage index is constructed based on shared bike order data and "electronic fence" spatial data to identify shared bike peak periods. Subsequently, a KNN-LightGBM model integrating spatiotemporal features is used to predict rental and return demand at these peak periods. The main steps include:

[0102] (1) Collect and preprocess shared bicycle orders and electronic fence data;

[0103] (2) Estimation of vehicle rental and return via electronic fence;

[0104] (3) Identify shared bicycle tidal points based on the vehicle shortage / parking shortage index;

[0105] (4) Extract the correlation information of adjacent electronic fences;

[0106] (5) Predicting tidal point rental demand based on the KNN-LightGBM model that integrates spatiotemporal features.

[0107] 2. Collect and preprocess shared bicycle order and geofence data. This includes the following processes: (1) removing outliers; (2) estimating the center point of the geofence and the number of parking spaces.

[0108] 3. Estimation of bike rental and return based on electronic fences. This includes the following steps: (1) Extracting the origin and destination location information of shared bike orders; (2) Matching the origin and destination of orders to electronic fences using the HNSW algorithm; (c) Calculating the rental and return demand for each electronic fence in each time period.

[0109] 4. Identifying shared bicycle tidal point based on missing bicycle / missing stop index. Including the following procedures (1) determining shared bicycle tidal point identification rule; (2) dividing electronic fence state type; (3) constructing missing bicycle / missing stop index; (4) calculating electronic fence missing bicycle / missing stop index; (5) identifying tidal point.

[0110] 5. Extracting correlation information of adjacent electronic fences. Including the following procedures (1) calculating correlation weight between electronic fences; (2) constructing vehicle rental and return time series matrix; (3) calculating correlation of adjacent electronic fences based on KNN algorithm.

[0111] 6. Predicting tidal point rental and return demand based on KNN-LightGBM model fusing spatio-temporal features. Including the following procedures (1) constructing and extracting key features affecting shared bicycle rental and return demand; (2) fusing model input features; (3) predicting tidal point rental and return demand based on LightGBM algorithm.

[0112] When identifying shared bicycle tide and predicting rental and return demand, the scheme provided by the present application and its preferred scheme has the following advantages compared with the traditional method: (1) considering the capacity of a single electronic fence, the missing bicycle / missing stop index can quickly obtain the tidal situation of each electronic fence and identify the tidal type (missing bicycle tidal point / missing stop tidal point) and spatio-temporal distribution. (2) At the same time, the influence of spatio-temporal features and correlation of adjacent electronic fences on rental and return demand prediction is considered, and a KNN-LightGBM rental and return demand prediction model based on fused spatio-temporal features is proposed, which effectively improves the prediction accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0113] The present application will be further described in detail below in combination with the drawings and specific embodiments:

[0114] Figure 1 The technical framework diagram for the shared bicycle tide point identification and rental and return demand prediction research of the embodiment of the present application.

[0115] Figure 2 The order OD point matching diagram of the embodiment of the present application.

[0116] Figure 3 The spatial distribution diagram of shared bicycle tidal point in the research area of the embodiment of the present application.

[0117] Figure 4 The result comparison diagram of the tidal point rental / return demand prediction based on KNN-LightGBM model in the embodiment of the present application under 1 hour and 4 hour time scales;

[0118] Among them, (a) 1 hour rental demand prediction result;

[0119] (b) 1 hour return demand prediction result;

[0120] (c) 4-hour rental demand prediction results;

[0121] (d) 4-hour drop-off demand prediction results. DETAILED DESCRIPTION

[0122] In order to make the features and advantages of the patent more apparent, the following specific examples are described in detail as follows:

[0123] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0124] It should be noted that the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and it should also be understood that when the terms "comprise" and / or "include" are used in the specification, they indicate the presence of a feature, step, operation, device, component, and / or combination thereof.

[0125] The purpose of the present application is to identify shared bicycle tidal points and predict rental and drop-off demand. The basic framework of the model is shown in Figure 1 , which includes the following steps:

[0126] First, collect and pre-process shared bicycle order and electronic fence data;

[0127] Collect and process shared bicycle order data and electronic fence space data. The collected order data and electronic fence data are processed, mainly including two parts:

[0128] (1) Eliminate outliers. Eliminate orders with excessively large or small trip distances and durations;

[0129] (2) Electronic fence center point and parking space number estimation. Calculate the area and center point coordinate of each electronic fence, and estimate the number of parking locations.

[0130] Second, electronic fence rental and drop-off vehicle estimation;

[0131] (1) Extract the starting and ending location information of shared bicycle orders;

[0132] (2) Match the order starting and ending points to the electronic fence using the hierarchical navigable small world graph (HNSW) algorithm, as shown in Figure 2 ;

[0133] (3) Calculate the rental and drop-off demand of each electronic fence at each time period.

[0134] Third step, identifying shared bicycle tidal point based on space-time model;

[0135] (1) Determine the shared bicycle tidal point identification rule. According to the basic principle of "electronic fence in a short time due to a large number of shared bicycle rent or return, resulting in no car to rent or no parking space", determine the tidal point identification rule;

[0136] (2) Construct the lack of car / parking index. According to the state of the electronic fence in a certain period of time, construct the lack of car / parking index;

[0137] (3) Identify the tidal point. According to the lack of car / parking index, identify the shared bicycle tidal point.

[0138] Fourth step, extract the correlation information of adjacent electronic fences

[0139] (1) Set the correlation weight between electronic fences. According to the spatial distance and threshold value between electronic fences, set the correlation weight between electronic fences;

[0140] (2) Construct the vehicle rental time series matrix. According to the historical rental order data of the electronic fence, establish the vehicle rental time series matrix;

[0141] (3) Calculate the correlation of adjacent electronic fences. Use KNN algorithm to calculate the correlation degree between target electronic fence and adjacent electronic fence in local range in historical rental demand.

[0142] Fifth step, predict the tidal point rental demand based on KNN-LightGBM model fused with space-time features

[0143] (1) Extract the key features affecting shared bicycle rental demand. Construct the space-time feature matrix, and use principal component analysis (PCA) to extract the key features;

[0144] (2) Construct model input features. Fuse the key features extracted by PCA and the correlation degree of adjacent electronic fences calculated by KNN as input features;

[0145] (3) Predict the tidal point rental demand. Based on the above input features, use LightGBM algorithm to predict the tidal point rental demand.

[0146] In the above steps, identifying shared bicycle tidal point based on lack of car / parking index, extracting correlation information of adjacent electronic fences and predicting tidal point rental demand based on KNN-LightGBM model fused with space-time features are the focus of the present application, which will be discussed in detail in the following specific embodiment introduction.

[0147] (1) Identify the shared bicycle tidal point based on the lack of car / parking index.

[0148] In order to quickly identify the shared bicycle tidal point in a specific period and analyze its tidal spatio-temporal characteristics, the embodiment proposes to construct the lack of bicycle / lack of parking index based on shared bicycle order data and "electronic fence" spatial data for identifying shared bicycle tidal point, including:

[0149] (a) Determine the shared bicycle tidal point identification rule

[0150] The shared bicycle tidal point is defined as: the electronic fence with no bicycle available for rent or no parking space available due to a large number of shared bicycle rentals or returns within a time T.

[0151] (b) Divide the electronic fence state type

[0152] ① No shared bicycle: a(i,j,k) = 0

[0153] ② Only a small number of shared bicycles: 0 < a(i,j,k) ≤ f × p(i)

[0154] ③ There are sufficient shared bicycles and parking spaces: p(i) × f < a(i,j,k) ≤ p(i) × g

[0155] ④ Only a small number of parking spaces: p(i) × g < a(i,j,k) ≤ p(i)

[0156] ⑤ Mild accumulation: 0 < b(i,j,k) ≤ p(i) × h

[0157] ⑥ Severe accumulation: p(i) × h < b(i,j,k)

[0158] Where: a(i,j,k) is the number of shared bicycles available for rent in the i-th electronic fence on the j-th day in the k-th time period; b(i,j,k) is the number of missing parking spaces in the i-th electronic fence on the j-th day in the k-th time period; p(i) is the total number of parking spaces in the i-th electronic fence; T, f, g, h are variable parameters.

[0159] (c) Construct the lack of bicycle / lack of parking index

[0160] The lack of bicycle index c(i,j,k) and the lack of parking index d(i,j,k) of the i-th electronic fence on the j-th day in the k-th time period are defined as shown in equations (1) and (2).

[0161]

[0162]

[0163] (d) Calculate the electronic fence lack of bicycle / lack of parking index

[0164] Based on the definition of step (c) above, the electronic fence lack of bicycle / lack of parking index is calculated as shown in equations (3), (4), (5) and (6):

[0165]

[0166]

[0167]

[0168]

[0169] C(i,j,k) represents the total bike shortage index of the ith electronic fence, and D(i,j,k) represents the total stop shortage index of the ith electronic fence; E(i,j,k) is used to represent the bike shortage index of the shared bicycle system in the jth day and the kth time period, and F(i,j,k) represents the stop shortage index of the shared bicycle system in the jth day and the kth time period.

[0170] (e) Identifying tidal points

[0171] According to the calculated bike / stop shortage index, the shared bicycle tidal point is identified.

[0172] (2) Extracting adjacent electronic fence correlation information

[0173] In order to obtain the correlation between adjacent electronic fences, the present application uses KNN algorithm to calculate the correlation degree of the target tidal point and the surrounding electronic fences in the historical rental and return demand:

[0174] (a) Setting the correlation weight between electronic fences

[0175] Considering the spatial distance between electronic fences, the correlation weight between electronic fences is represented by D(x a , x b ), as shown in formula (7).

[0176]

[0177] dist(x a ,x b ) represents the distance between electronic fences x a and x b ; R is the distance threshold value. When the distance between two electronic fences is less than or equal to the size of the threshold value R, the correlation weight coefficient of the two electronic fences is 1, otherwise it is 0.

[0178] (b) Constructing a vehicle rental and return time series matrix

[0179] A time series matrix X is established by the historical rental data of the electronic fence, as shown in formula (8), x a t represents the number of rentals of electronic fence x a at t period, and similarly, a return demand time series matrix is constructed according to the same rule.

[0180]

[0181] (c) Calculate the correlation of adjacent electronic fences based on KNN

[0182] Calculate the correlation of electronic fences with different parameters K. Electronic fence x a and x b The similarity score Sim(x a , x b ) is calculated by formula (9) to obtain the electronic fence most similar to the tidal point. Finally, the evaluation indicators of the prediction model: root mean square error (RMSE), mean absolute error (MAE), mean determination coefficient (R 2 ) and correlation coefficient (r) are used to determine the optimal value of parameter K, as shown in formulas (10), (11), (12) and (13).

[0183]

[0184]

[0185]

[0186]

[0187]

[0188] (3) KNN-LightGBM model based on fusion of spatio-temporal features to predict tidal point rental demand

[0189] In order to obtain the spatio-temporal features and the correlation information of adjacent electronic fences that affect the rental demand of shared bicycles, the present application proposes a KNN-LightGBM rental demand prediction model based on the fusion of spatio-temporal features.

[0190] (a) Construct and extract key features affecting shared bicycle rental demand

[0191] The influence of spatio-temporal data such as shared bicycle orders, city information points (POI), roads, population, land use, temperature, wind speed, etc. on rental demand is considered comprehensively, and the spatio-temporal features of tidal points are analyzed, and the PCA algorithm of SPSS software is used to extract z main components.

[0192] (b) Construct model input features

[0193] The feature vectors extracted by PCA and the correlation information of adjacent electronic fences calculated by KNN are fused as input features.

[0194] (c) Predict tidal point rental demand based on LightGBM algorithm

[0195] ① Construct the training set.

[0196] The training set is constructed using the shared bicycle rental and return history data, and the features obtained by the construction are input into the LightGBM algorithm.

[0197] ② Train and optimize the LightGBM model.

[0198] First, linearly combine M weak regression trees into a strong regression tree, as shown in formula (14).

[0199]

[0200] Where F(x) is the sum of the final output values, f m (x) is the output value of the mth weak regression tree.

[0201] Then use the histogram algorithm to combine the input features. The basic idea is to first divide the continuous feature values into K integers, and construct a histogram with a width of K. When traversing the data, the discretized value is used as an index and accumulated in the histogram to find the best decision tree split point.

[0202] Finally, use the leaf-wise leaf growth strategy with depth limit to find the leaf with the maximum split gain from all the current leaves for splitting, and set the maximum depth limit.

[0203] ③ Predict the tidal point rental demand. Finally, predict and output the tidal point rental demand.

[0204] In the specific test example of the application, based on the shared bicycle order data and electronic fence data of a city in China, the effectiveness of the application in identifying shared bicycle tidal points and predicting rental demand is analyzed.

[0205] In the test, in the lack of car / parking space index for identifying tidal points, the time interval T in the definition of tidal point is set to 1 hour, and the parameters f, g, h for classifying electronic fence states are set to 0.2, 0.8, and 0.2 respectively. The classified electronic fence states are as follows:

[0206] ① No shared bicycles: a(i,j,k)=0

[0207] ② Only a small amount of shared bicycles: 0<a(i,j,k)≤0.2×p(i)

[0208] ③ There are sufficient shared bicycles and parking spaces: p(i)×0.2<a(i,j,k)≤p(i)×0.8

[0209] ④ Only a small amount of parking spaces: p(i)×0.8<a(i,j,k)≤p(i)

[0210] • Mild congestion: 0 < b(i,j,k) < p(i) x 0.2

[0211] • Severe congestion: p(i) x 0.2 < b(i,j,k)

[0212] The spatial distribution of the 20 tidal points with the largest total car shortage index C(i,j,k) and total stop shortage index D(i,j,k) is shown in FIG. 2. Figure 3

[0213] In predicting the demand for rental and return at the tidal point, 8 key features were extracted using the PCA algorithm. The correlation information of the adjacent electronic fences was extracted, the distance threshold was set to 500 meters, and the nearest neighbor sample parameter was set to 6. The time step of the rental and return demand prediction was set to 1 hour and 4 hours. In the training of the prediction model, the learning rate of the KNN-LightGBM rental and return demand prediction model based on the fusion of spatio-temporal features was set to 0.1, the number of leaf nodes was set to 8, and the maximum depth of the tree model was set to 6. The results are shown in Table 1. Figure 4 The root mean square error of the model proposed in the present application is 1.14, the mean absolute error is 0.67, and the average determination coefficient and correlation coefficient are both 0.84. Compared with the LightGBM model, the root mean square error and the mean absolute error are reduced by 10% and 11%, respectively, and the average determination coefficient and correlation coefficient are increased by 3% and 4%, respectively. Therefore, the present application considers the spatio-temporal features and the spatial correlation of the electronic fence, and proposes a KNN-LightGBM rental and return demand prediction model based on the fusion of spatio-temporal features, which effectively improves the prediction accuracy.

[0214] Table 1 Comparison of rental and return demand prediction results of different models

[0215]

[0216] Note: a is 1 hour rental demand prediction, b is 1 hour return demand prediction, c is 4 hour rental demand prediction, and d is 4 hour return demand prediction.

[0217] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage media, etc.) having computer-usable program code embodied in the medium.

[0218] ​The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps in one or more flow or blocks.

[0219] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps in one or more flow or blocks.

[0220] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps in one or more flow or blocks.

[0221] The above description is only preferred embodiments of the present application, and is not intended to limit the present application to other forms described above. Any person skilled in the art can make changes or modifications to the above embodiments without departing from the technical content of the present application. Any simple modification, equivalent change and modification of the above embodiments according to the technical essence of the present application still belongs to the protection scope of the present application.

[0222] The present application is not limited to the above best mode, and anyone can derive other various forms of shared bicycle tide point identification and rental demand prediction methods under the inspiration of the present application. Any equivalent change and modification made according to the scope of the present application shall be covered by the present application.

Claims

1. A method for identifying shared bicycle tidal points and predicting rental and return demand, characterized in that, Includes the following steps: Step S1: Collect and preprocess shared bicycle orders and geofence data; Step S2: Estimation of vehicle rental and return via electronic fence; Step S3: Identify shared bicycle tidal points based on a spatiotemporal model; Step S3 specifically includes the following steps: Step S31: Determine the rules for identifying shared bicycle tidal points: Determine the rules for identifying tidal points based on the basic principle of "electronic fences where there are no bicycles available for rent or no parking spaces available due to a large number of shared bicycles being rented or returned in a short period of time". Step S32: Construct the vehicle / parking shortage index: Based on the status of the electronic fence during a certain period of time, construct the vehicle / parking shortage index; Step S33: Identify tidal points: Identify shared bicycle tidal points based on the bike shortage / parking index; Step S4: Extract correlation information between adjacent electronic fences; Step S5: Predict tidal point rental and return demand based on the KNN-LightGBM model that integrates spatiotemporal features; Step S5 specifically includes the following steps: Step S51: Extract key features affecting the demand for shared bicycle rentals: Construct a spatiotemporal feature matrix and use principal component analysis (PCA) to extract key features; Step S51 is as follows: The study comprehensively considers at least the impact of spatiotemporal data on shared bicycle orders, urban information points, roads, population, land use, temperature, and wind speed on rental demand. It also analyzes the spatiotemporal characteristics of tidal points and uses the PCA algorithm of SPSS software to extract z principal components. Step S52: Construct model input features: Combine the key features extracted by PCA and the correlation between adjacent electronic fences calculated by KNN as input features; Step S53: Predict tidal point rental demand: Based on the input features, predict tidal point rental demand using the LightGBM algorithm: Step S53 specifically includes: Step S531: Construct the training set; A training set was constructed using historical data of shared bicycle rentals, and the resulting features were input into the LightGBM algorithm. Step S532: Train and optimize the LightGBM model: First, the M weak regression trees are linearly combined into a strong regression tree, as shown in formula (14): (14) Where F(x) is the sum of the final output values, f m (x) is the output value of the m-th weak regression tree; Then, the input features are merged using the histogram algorithm: the continuous feature values ​​are divided into K integers and a histogram of width K is constructed. When traversing the data, the discretized values ​​are used as indices and accumulated in the histogram to find the best decision tree split point. Finally, a leaf-wise growth strategy with depth restrictions is used to find the leaf with the maximum splitting gain among all the current leaves and split it, and a maximum depth restriction is set. Step S533: Predict and output the rental and return demand at tidal points.

2. The method for identifying shared bicycle tidal points and predicting rental demand according to claim 1, characterized in that, In step S1, the collected order data and geofence data are processed, specifically including the following steps: Step S11: Remove outliers: Remove orders with excessively large or small travel distances and durations; Step S12: Estimation of the center point and number of parking spaces of the electronic fence: Calculate the area and center point coordinates of each electronic fence, and assess the number of parking spaces.

3. The method for identifying shared bicycle tidal points and predicting rental demand according to claim 1, characterized in that, Step S2 specifically includes the following steps: Step S21: Extract the origin and destination location information of shared bicycle orders; Step S22: Use the hierarchical navigable small-world graph algorithm to match the order's origin and destination to the electronic fence; Step S23: Calculate the rental and return demand for each electronic fence in each time period.

4. The method for identifying shared bicycle tidal points and predicting rental demand according to claim 1, characterized in that, Step S4 specifically includes the following steps: Step S41: Set the relevant weights between electronic fences: Set the relevant weights between electronic fences based on the spatial distance and threshold between them; Step S42: Construct a vehicle rental and return time series matrix: Construct a vehicle rental and return time series matrix based on historical rental and return order data from the geofence; Step S43: Calculate the correlation between adjacent electronic fences: Use the KNN algorithm to calculate the correlation between the target electronic fence and neighboring electronic fences in terms of historical rental demand within a local area.

5. The method for identifying shared bicycle tidal points and predicting rental demand according to claim 1, characterized in that, In step S33, identifying shared bicycle tidal points based on the bike shortage / parking shortage index specifically includes: Step S331: Determine the rules for identifying shared bicycle tidal points The shared bicycle tidal point is defined as: the electronic fence within time T where there are no bicycles available for rent or no parking spaces available due to a large number of shared bicycles being rented or returned. Step S32: Classify the electronic fence status type ① No shared bicycles: a(i,j,k) = 0 ② Only a small number of shared bicycles: 0 < a(i,j,k) ≤ f × ​​p(i) ③ Sufficient shared bicycles and parking spaces are available: p(i) × f < a(i,j,k) ≤ p(i) × g ④ Only a limited number of parking spaces: p(i) × g < a(i,j,k) ≤ p(i) ⑤ Slightly stacked: 0 < b(i,j,k) ≤ p(i) × h ⑥ Heavy stacking: p(i) × h < b(i,j,k) Where: a(i,j,k) is the number of shared bicycles available for rent on day j and time period k of the i-th electronic fence; b(i,jk) is the number of parking spaces missing on day j and time period k of the i-th electronic fence; p(i) is the total number of parking spaces in the i-th electronic fence; T, f, g, and h are variable parameters; Step S333: Construct the vehicle shortage / parking shortage index The vehicle shortage index c(i,j,k) and parking shortage index d(i,j,k) for the j-th day and k-th time period of the i-th electronic fence are defined as shown in equations (1) and (2): (1) (2) Step S334: Calculate the electronic fence vehicle / parking shortage index Based on the definition in step S333, the electronic fence missing vehicle / parking index is calculated as shown in equations (3), (4), (5) and (6): (3) (4) (5) (6) C(i,j,k) represents the total bike shortage index of the i-th electronic fence, and D(i,j,k) represents the total parking shortage index of the i-th electronic fence; E(i,j,k) represents the bike-sharing system's performance on day j. The bike shortage index for a time period, F(i,j,k) represents the bike shortage index of the shared bike system in the k-th time period on the j-th day; Step S335: Identify tidal points The shared bicycle shortage / parking index is used to identify the peak times for shared bicycles.

6. The method for identifying shared bicycle tidal points and predicting rental demand according to claim 4, characterized in that, In step S4, Step S41 is as follows: Considering the spatial distance between electronic fences, use D(x) a x b The weights between the electronic fences are represented by , as shown in equation (7): (7) dist(x a ,x b ) represents the electronic fence x a and x b The distance between the two electronic fences; R is the distance threshold; when the distance between the two electronic fences is less than or equal to the threshold R, the correlation weight coefficient between the two electronic fences is 1, otherwise it is 0; Step S42 is as follows: A time series matrix X is constructed using historical car rental data from electronic fences, as shown in equation (8). a t Indicates electronic fence x a The number of cars rented in time period t, and a time series matrix of car return demand constructed according to the same rules: (8) Step S43 is as follows: Electronic fence x a and x b Similarity score Sim(x) a x b The calculation formula is Equation (9), which yields the electronic fence most similar to the tidal point; finally, the evaluation indicators of the prediction model are: root mean square error (RMSE), mean absolute error (MAE), and mean coefficient of determination (R²). 2 And the correlation coefficient r, as shown in formulas (10), (11), (12) and (13), determine the optimal value of parameter K: (9) (10) (11) (12) (13)。

Images