A shared bicycle station differentiated accurate delivery method based on an extreme learning machine

Abstract

This invention relates to the fields of deep learning and public transportation technology, and discloses a method for differentiated and precise deployment of shared bicycles at various stations based on Extreme Learning Machine (ELM). First, an ELM prediction model is constructed. Then, data on the number of shared bicycles at each station in the target area within a specified time interval is collected, and the actual usage of shared bicycles at each station within that time interval is calculated to form a historical dataset. The historical dataset from day t (N+1 days prior) is normalized and formatted into a three-dimensional array {N+1,7,1}, which is then input into the ELM prediction model for training to predict the demand for shared bicycles at each station on day t+1 within the corresponding time interval. When the total predicted demand exceeds the threshold of the total number of schedulable bicycles in the target area, the total number to be reduced is calculated, resulting in the actual number of shared bicycles deployed at each station. This invention proposes using a deep learning model to accurately predict the demand for shared bicycles at each station from a time perspective, thereby scientifically guiding vehicle deployment.

Description

Technical Field

[0001] This invention relates to the fields of deep learning and public transportation technology, specifically to a method for differentiated and precise deployment of shared bicycles at designated stations based on Extreme Learning Machine. Background Technology

[0002] Currently, commuting options are becoming increasingly diverse and convenient. However, public transportation modes such as subways, buses, and trams suffer from shortcomings in last-mile connectivity, failing to provide truly door-to-door service. While ride-hailing services like Didi and taxis offer door-to-door service, they are economically costly for the average person and frequent travelers. The emergence of shared bicycles has precisely solved the "last mile" problem, allowing users to ride from public transportation stops to their destinations, effectively filling the last-mile transportation gap.

[0003] However, in densely populated areas such as public transportation hubs and commercial districts, the number of shared bikes deployed often fails to meet passenger travel demand during peak hours; while in sparsely populated areas, the number of bikes deployed often exceeds actual demand, resulting in idle and wasted vehicle resources. Meanwhile, station demand is dynamically affected by various factors such as time of day, traffic control policies, commercial relocation, and weather. Therefore, for shared bike operators, optimizing deployment strategies is a core operational challenge, specifically including key issues such as deployment site planning and quantity control. In conventional deployment models, operators tend to concentrate deployments in high-demand areas such as subway and bus stations, popular commercial districts, and markets. However, even with concentrated deployments during rush hours, it is still difficult to meet the surge in travel demand, leading to a supply-demand imbalance where users struggle to find bikes.

[0004] The application of machine learning technology in the transportation sector is booming, particularly in areas such as traffic flow statistics and real-time road condition analysis, where numerous research findings have been achieved. By mining and analyzing massive amounts of historical traffic data through machine learning algorithms, patterns and trends in traffic flow changes can be accurately identified, providing data support and decision-making basis for road management and transportation planning.

[0005] Deep learning methods have distinguished themselves among machine learning algorithms due to their core advantages, such as automatic feature extraction, compatibility with linear and nonlinear problems, multi-task learning, and large-scale data processing. Their widespread application in the transportation sector and significant potential in traffic flow prediction have made them a preferred solution in scientific research and engineering. For example, Long Short-Term Memory (LSTM) networks possess excellent long-term time-series memory capabilities, and Convolutional Neural Networks (CNNs) excel at feature extraction; both have been widely used in traffic flow prediction scenarios.

[0006] However, deep learning models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Long Short-Term Memory Networks (LSTMs) generally suffer from high internal network complexity. In practical applications, to ensure prediction and classification performance, multi-layered stacked architectures are typically constructed, leading to redundancy in knowledge transfer and feature interaction processes. This not only increases the consumption of computing resources but also prolongs the model training cycle, reduces real-time response performance, and makes it difficult to adapt to dynamic demand scenarios. It also fails to effectively solve the core problem of shared bicycle operators' inability to accurately control the deployment volume at different times and locations.

[0007] Extreme Learning Machines (ELMs) feature a simple network topology. Knowledge transfer between neurons does not rely on iterative optimization algorithms such as gradient descent; instead, it simply generates hidden layer parameters randomly and directly solves for the optimal output weights using the Moore-Penrose formula. This characteristic frees ELMs from the constraints of complex iterative training, significantly improving training speed and execution efficiency while maintaining excellent generalization ability. It is well-suited for engineering tasks such as predicting demand for shared bicycles. Summary of the Invention

[0008] To address the aforementioned problems, the purpose of this invention is to provide a differentiated and precise deployment method for shared bicycle stations based on Extreme Learning Machine (ELM). This method uses a deep learning model to accurately predict the demand for shared bicycles at each station over time, thereby scientifically guiding vehicle deployment. The technical solution is as follows:

[0009] A method for differentiated and precise deployment of shared bicycles at designated stations based on extreme learning machine includes the following steps:

[0010] Step 1: Constructing the Extreme Learning Machine Prediction Model: The Extreme Learning Machine prediction model is built using the Python Hpelm library. The prediction model structure includes an input layer, an output layer, and two hidden layers. The first hidden layer is used to extract non-linear features between data, and the second hidden layer is used to extract linear features between data.

[0011] Step 2: Construct a shared bicycle deployment algorithm based on Extreme Learning Machine:

[0012] Step 2.1: Collect data on the number of shared bicycles at each station in the target area within a specified time interval using positioning technology, calculate the actual usage of shared bicycles at the stations within that time interval, and form a historical dataset;

[0013] Step 2.2: Normalize the historical dataset of N+1 days before day t and organize it into a three-dimensional array format. Input the data into the Extreme Learning Machine prediction model for training, so as to predict the demand for shared bicycles at each site in the corresponding time interval on day t+1.

[0014] Step 2.3: When the total demand for shared bicycles at each station in the target area exceeds the threshold of the total number of schedulable bicycles in the target area, calculate the total amount to be reduced, and allocate the reduction amount to each station according to the ratio coefficient of the initial predicted demand to the total predicted demand, and finally obtain the actual number of shared bicycles deployed at each station.

[0015] The beneficial effects of this invention are as follows: This invention proposes the use of an Extreme Learning Machine (ELM) model. The ELM network structure is simple, and network parameters are derived through formulas rather than iterative derivations during model training. Its prediction accuracy is comparable to traditional networks such as CNN and LSTM, but its prediction speed is tens of times faster than CNN and LSTM. Therefore, using the ELM model allows for accurate prediction and deployment of shared bicycles at various stations from a time perspective, dynamically meeting passenger needs and avoiding resource waste. Attached Figure Description

[0016] Figure 1 This is the structure of the Extreme Learning Machine prediction model of the present invention.

[0017] Figure 2 This is a schematic diagram of a shared bicycle deployment scheme based on ELM. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0019] This invention proposes a differentiated and precise deployment method for shared bicycle stations based on extreme learning machine, the technical solution of which is as follows:

[0020] Step 1: Construct the ELM prediction model:

[0021] This paper uses the Python Hpelm library to build an Extreme Learning Machine (ELM) prediction model. As an open-source library, Hpelm not only implements the basic single-hidden-layer feedforward ELM, but also supports various ELM variants, which can meet the modeling needs of various scenarios such as linear / nonlinear. At the same time, it supports seamless switching between regression and classification tasks, perfectly matching the core characteristics of the ELM algorithm: no iterative training required and fast convergence.

[0022] The structural design of the extreme learning machine prediction model of this invention is as follows: Figure 1As shown, there are only four simple layers. The first and last layers are the input and output layers, respectively, and the second and third layers are hidden layers. The second layer uses 200 hidden layer neurons and employs the sigmoid activation function to extract non-linear features between data. The third layer uses 100 hidden layer neurons and employs the linear activation function to extract linear features between data.

[0023] Step 2: ELM-based algorithm for shared bicycle deployment:

[0024] ELM-based shared bicycle deployment solutions, such as Figure 2 As shown, the manufacturer's control center uses deep learning algorithms to predict the demand for shared bicycles at each site. After completing the scheduling analysis in conjunction with preset deployment rules, it coordinates the deployment of shared bicycles at each site by the maintenance personnel.

[0025] Domestic bike-sharing companies generally adopt a two-tiered model of "city-level zoned management + refined management of stations within the region" for deployment and operation. Within the designated city operation zones, a dedicated operations and maintenance team is responsible for vehicle dispatching, maintenance, and clearing of illegally parked vehicles. The regions are further subdivided into shared bike stations.

[0026] Within the same operational zone, it is necessary to accurately track the usage of shared bicycles at each station at different times. From a demand correlation perspective, the demand for shared bicycles is positively correlated with regional passenger traffic, but the two are not a fixed ratio—demand is affected by multiple factors such as time-of-day rhythms (e.g., morning and evening rush hours, off-peak hours), traffic control policies (e.g., traffic restrictions, temporary controls), commercial relocation (e.g., opening / closing of shopping districts), and weather conditions (e.g., rain, high temperatures), exhibiting dynamic fluctuations. Therefore, the actual demand at each station must be determined primarily based on real-time shared bicycle usage data.

[0027] Regional passenger traffic is not equivalent to the demand for shared bicycles. Currently, shared bicycle terminals, through the integration of positioning chips, can achieve real-time location tracking of shared bicycles. This helps determine whether a user's parking location is within a pre-defined shared bicycle planning station area, or provides users with information on station bicycle distribution, facilitating real-time queries about the number of available shared bicycles at each station. Based on the shared bicycle location information collected using the aforementioned positioning technology, further statistical analysis yields shared bicycle usage status data for each station within a specified time period.

[0028] The logic for collecting station vehicle data based on positioning technology is as follows: Assuming that the data is collected over the time interval of day t... Shared bike usage at site A (The time interval for statistical data excludes the time period during which maintenance personnel dispatch vehicles.) .exist The number of shared bicycles of a certain brand at station A is determined using location technology at all times. ,exist The number of vehicles of this brand at this site is obtained in real time. ,in This includes vehicles stored at this station and vehicles ridden from other stations to this station.

[0029] Therefore in The actual usage of shared bicycles at time station A is and The absolute value of the difference, that is:

[0030] ;

[0031] in, There are two possible numerical states: or .like This indicates that the site is If all bikes are ridden away during a given time period and no external bikes enter the site, the shared bikes at the station are likely in short supply. In this situation, the manufacturer can determine whether to deploy Nx additional shared bikes. The manufacturer dynamically adjusts Nx based on factors such as the actual number of available bikes.

[0032] The original dataset for the model was normalized using the Z-score normalization method. Data normalization scales datasets with large variations or high numerical values ​​to a reasonable range according to uniform rules, which helps improve the accuracy of model predictions. The normalization formula is as follows:

[0033] ;

[0034] in, For the raw data, Z is called the standard score. The mean of the dataset. denoted as the standard deviation of the dataset. After normalization, the dataset is screened for short-term data whose standard score (Z-value) is greater than 3 or less than -3. If such data exists, it is identified as an outlier and replaced with the dataset mean to improve data quality.

[0035] In the demand forecasting model construction phase, considering the significant impact of cyclical changes in demand during weekday morning rush hours, a sliding window of length 7 and step size 1 is used to continuously sample the normalized dataset and sequentially input it into the neural network for training. The corresponding training input data dimension representation of the neural network is (N+1,7,1). The short-term partitioning method with a length of 7 can quickly respond to sudden changes in demand caused by factors such as mall relocation and temporary control measures, avoiding the lag caused by long-term data.

[0036] Therefore, the input training set for ELM-based deep learning algorithms is... , The training set is then rearranged into a three-dimensional array of shape (N+1,7,1) and input into the ELM prediction model for training and prediction. The prediction results are then denormalized to obtain... That is, on day t+1, site A is The demand for shared bicycles at any given time.

[0037] In practice, the number of shared bikes that can be dispatched in a region is limited. When the total demand for shared bikes at each station in the region, as predicted by the ELM prediction model, exceeds the number that can be dispatched, the number of shared bikes deployed at each station needs to be adjusted.

[0038] In the scenario of shared bicycle area dispatching, the total number of dispatchable bicycles in the area is set as a threshold. When the sum of the demand for all stations on day t+1 generated by the Extreme Learning Machine (ELM) prediction model exceeds the threshold, i.e.:

[0039] ;

[0040] Adjustments were made to the amount of material distributed at each site.

[0041] The total number of shared bicycles that need to be reduced is calculated as follows:

[0042] ;

[0043] The initial forecast demand percentages for each site are as follows:

[0044] ;

[0045] The weight reduction for each site is the initial percentage, and the total weight reduction is:

[0046] ;

[0047] The number of shared bicycles to be reduced at each site is as follows (non-integer reductions should be rounded):

[0048] ;

[0049] Ultimately, the actual number of shared bicycles allocated to each station was:

[0050] ;

[0051] This involves calculating the percentage of each site's predicted demand relative to the total predicted demand, and then comparing the total predicted demand with a threshold. The difference is allocated to each site according to the proportion coefficient to obtain the reduction amount. The non-integer reduction amount is rounded to determine the final delivery amount.

[0052] Suppose that the historical dataset of shared bicycle usage collected from bus stop A on day t for the previous N+1 days at time T is {…, 18, 13, 17, 11, 40, 35, 14}, with a mean of 15 and an overall standard deviation of 6. Among these, {40, 35} represent short-term outliers caused by a two-day promotional event at a shopping mall near stop A. Manufacturers typically cannot effectively predict such sudden events or lack corresponding shared bicycle management and dynamic scheduling capabilities. After normalizing the original dataset, it becomes clear that the standard scores (Z-scores) for these two days are both greater than 3. If the original dataset containing these outliers is directly used to train the model and predict the shared bicycle demand on day t+1, the prediction will deviate significantly from the actual demand. Therefore, replacing the outlier data with the dataset mean before training and prediction yields the most accurate predicted shared bicycle demand for stop A on day t+1 when there are no sudden events.

[0053] Compared to a fixed-location, fixed-quantity bike deployment strategy, the ELM-based bike-sharing deployment solution can deploy bikes according to predicted demand, dynamically meeting passenger needs and avoiding resource waste.

Claims

1. A method for differentiated and precise deployment of shared bicycles at designated stations based on extreme learning machine, characterized in that, Includes the following steps: Step 1: Constructing the Extreme Learning Machine Prediction Model: The Extreme Learning Machine prediction model is built using the Python Hpelm library. The prediction model structure includes an input layer, an output layer, and two hidden layers. The first hidden layer is used to extract non-linear features between data, and the second hidden layer is used to extract linear features between data. Step 2: Construct a shared bicycle deployment algorithm based on Extreme Learning Machine: Step 2.1: Collect data on the number of shared bicycles at each station in the target area within a specified time interval using positioning technology, calculate the actual usage of shared bicycles at the stations within that time interval, and form a historical dataset; Step 2.2: Normalize the historical dataset of N+1 days before day t and organize it into a three-dimensional array format. Input the data into the Extreme Learning Machine prediction model for training, so as to predict the demand for shared bicycles at each site in the corresponding time interval on day t+1. Step 2.3: When the total demand for shared bicycles at each station in the target area exceeds the threshold of the total number of schedulable bicycles in the target area, calculate the total amount to be reduced, and allocate the reduction amount to each station according to the ratio coefficient of the initial predicted demand to the total predicted demand, and finally obtain the actual number of shared bicycles deployed at each station.

2. The method for differentiated and precise deployment of shared bicycle stations based on extreme learning machine according to claim 1, characterized in that, The first hidden layer uses 200 hidden layer neurons and employs the sigmoid activation function; the second hidden layer uses 100 hidden layer neurons and employs the linear activation function.

3. The method for differentiated and precise deployment of shared bicycle stations based on extreme learning machine according to claim 1, characterized in that, Step 2.1 specifically involves: Statistical analysis of the time interval on day t Shared bike usage at site A ;exist The system uses location technology to determine the current number of shared bicycles of a specific brand at station A. ,exist The number of vehicles of the brand mentioned at site A is obtained in real time. ,in This includes vehicles stored at this station and vehicles ridden from other stations to this station; Then time interval Actual usage of shared bicycles at site A for and The absolute value of the difference, that is: ； This ultimately forms a historical dataset.

4. The method for differentiated and precise deployment of shared bicycle stations based on extreme learning machine according to claim 3, characterized in that, Step 2.2 specifically involves: Input training set is The training set is then normalized and rearranged into a three-dimensional array of shape (N+1, 7, 1), which is then input into the Extreme Learning Machine prediction model for training and prediction. That is, on day t+1, site A is The demand for shared bicycles at any given time; N+1 is the number of days corresponding to the data in the training set, 7 is the length of the sampling window, and 1 is the sampling step size.

5. The method for differentiated and precise deployment of shared bicycle stations based on extreme learning machine according to claim 3, characterized in that, Step 2.3 specifically involves: Set the total number of schedulable bicycles in the target area as a threshold. When the total demand for all sites on day t+1, generated by the extreme learning machine prediction model, exceeds a threshold, the delivery volume for each site is adjusted, i.e.: ； in, The demand for i-bikes at the station; Represents the collection of all sites; The total number of shared bicycles needs to be reduced. for: ； Initial forecast demand percentage for each site for: ； The weight reduction for each site is the initial percentage, and the total weight reduction is... for: ； Each station should reduce the number of shared bicycles. for: ； Ultimately, the actual number of shared bicycles allocated to each station for: 。

Images