• No results found

Design and Research of Bike sharing Delivery Forecasting

N/A
N/A
Protected

Academic year: 2020

Share "Design and Research of Bike sharing Delivery Forecasting"

Copied!
7
0
0

Loading.... (view fulltext now)

Full text

(1)

2018 3rd International Conference on Information Technology and Industrial Automation (ICITIA 2018) ISBN: 978-1-60595-607-7

Design and Research of Bike-sharing Delivery

Forecasting

Tangchao Li and Zhenrong Zhang

ABSTRACT

As the bake sharing is popular all over the world, the problems they display are becoming more and more serious. The excessive release of bicycles, random placement, occupation of sidewalks, motor vehicle lanes, etc., cause congestion and even traffic accidents. For the irrationality of bicycle delivery, this paper proposes to use neural network prediction method to predict the amount of each parking spot, which can solve the problem of excessive delivery. The neural network model used can also overcome the shortcomings of traditional regression model

INTRODUCTION

The bake sharing model is borrowed from the public bicycle system. The bicycle transportation mode saves transportation energy and land consumption, can effectively reduce urban traffic pollution, alleviate traffic congestion, and has low personal travel expenses. It is a healthy low-carbon travel mode integrating transportation, fitness and entertainment, and is suitable for all income classes. Residents use. In addition to the above advantages, the bicycle transportation mode is the most accessible means of transportation in addition to the walking mode. The establishment of the urban public bicycle system can effectively solve the problem of "last 1km" of urban residents' bus travel. At present, the shared bicycles are mainly short-distance and connecting traffic used in a short period of time, and are suitable for deploying parking spots in hubs, residential areas, commercial areas,

Tangchao Li, Zhenrong Zhang, School of Computer, Electronics and Information, Guangxi University, Nanning, China

(2)

tourist attractions and other passenger flow distribution centers. However, the survey found that the number of deliveries in many areas is not reasonable. The unscientific quantity of the delivery caused the bicycles in some rental points to be in short supply, and was often borrowed out. There were also many bicycle rentals at the rental point, which wasted resources [1].

Based on the detailed statistics of the daily bicycle lending situation at the parking spot, the land use situation of the parking area, the population, the user distance and the coverage area data, the artificial neural network model is established to solve the number of delivery, and finally the parking point. Reasonable forecasting of the number of deliveries.

VOLUME FORECAST

Volume Factor Analysis

The forecast of bike sharing delivery is affected by many factors. The following are the main considerations in this paper.

The forecast of the amount of delivery needs to consider the number of people and the type of land used nearby; for example, shopping malls, hospitals, residential areas and other crowded areas, the number of placements can be appropriately increased, while areas such as suburbs, old towns, etc. Reduce the amount of delivery in moderation.

The distance between the placement point and the user. There may be multiple parking spots around the same user. The different distances may cause the user to select different parking points. Therefore, it is necessary to reasonably allocate the delivery quantity according to the distance between each delivery point and the user. According to the coverage area of the bicycle parking spot, if the coverage area of the delivery point is relatively large, the number of delivery needs to be increased to meet the needs of users in the area.

Finally, the location planning of shared bicycle parking spots needs to be combined with the urban public transportation system such as subway and public transportation, and the urban public transportation can be closely combined to completely solve the problem of “last mile” in the citizens' travel.

Model Analysis

(3)

resulting in the possibility of distortion after the model is established, the method has great limitations. The other type is time series prediction. This kind of method does not need to consider various influencing factors, but builds the public bicycle demand forecasting model of the site by extracting the law of historical data changes, and then speculates the future makeup data of the site. Time series prediction methods can be generally divided into three categories; traditional linear time series prediction method, traditional nonlinear time series prediction method and neural network time series prediction method. The traditional time series prediction method is based on one-dimensional space, and the model is simple, but it is difficult to deal with complex situations in practical applications. The neural network-based time series prediction method is suitable for prediction with time-varying, nonlinear and uncertainty due to its self-organizing self-learning ability [2].

NEURAL NETWORKS

A neural network is a multi-layer neural network, which can be generally divided into an input layer, an implicit layer, and an output layer, wherein the hidden layer can be one or more layers. The neurons between the upper and lower layers are fully connected, that is, each unit of the lower layer is connected to each unit of the upper layer, and there is no connection between each layer of neurons [3]. A typical three-layer artificial neural networks topology is shown in Figure 1.

...

...

...

...

...

...

X1

X2

Xm

...

Y1

Y2

Yn

Input layer Hidden layer Output layer

O1

O2

Ok

(4)

Suppose Xi is the input layer node, Oj is the hidden layer output node, Yk is the output layer node, and wij is the weight of the connection node i to the node j. Then

the implicit layer node output model is:

𝑂𝑗 = 𝑓(𝑤𝑖𝑗∗ 𝑥𝑖) (1)

The output layer node output model is:

𝑌𝑘= 𝑓(𝑤𝑗𝑘∗ 𝑂𝑗) (2)

Where f(*) is the activation function, the Relu function is used here, because the sigmoid function is easily oversaturated and the gradient disappears [4-5]. The function of Relu can be expressed as:

𝑓(𝑥) = { 0 , 𝑥 ≤ 0 𝑥, 𝑥 > 0 (3)

Error reverse propagation is the most prominent feature of artificial neural network algorithms, and its error is often calculated by the mean square error of the output and input. In addition, functions such as continuous micro-error function can be used to calculate the error, but if other error functions are used, the system calculation will be more complicated, so the error function should be used with caution. The mean squared function is defined as follows:

𝐸 =12∑ (𝑌𝑘 𝑘− 𝑦𝑘)2 (4)

The sum of the error generated by the sample training output and the input in the E artificial neural network, and yk is the actual value of the kth output node. From the above formula, we can easily find that each neuron node in the output layer will have an impact on the overall error. According to the error calculation result, according to the idea of reverse propagation, the weight is modified in the opposite direction. Generally, the steepest descent method (also called the gradient descent method) is used to operate. First, the local gradient between the hidden layer and the output layer is calculated as follows:

𝛻 =𝜕𝑤𝜕𝐸

𝑗𝑘 =

𝜕𝐸

𝜕 ∑(𝑤𝑗𝑘𝑂𝑗)

𝜕 ∑(𝑤𝑗𝑘𝑂𝑗)

𝜕𝑤𝑗𝑘 =

𝜕𝐸

𝜕 ∑(𝑤𝑗𝑘𝑂𝑗)𝑂𝑗 (5)

And

𝜕 ∑(𝑤𝜕𝐸

𝑗𝑘𝑂𝑗) =

𝜕𝐸

𝜕𝑌𝑘

𝜕𝑌𝑘

𝜕 ∑(𝑤𝑗𝑘𝑂𝑗) =

𝜕𝐸

(5)

So you can get

𝛻 =𝜕𝑤𝜕𝐸

𝑗𝑘= 𝑌𝑘(1 − 𝑌𝑘)(𝑌𝑘− 𝑦𝑘)𝑂𝑗 (7)

As can be seen from the above equation, in order to correct the value of Yk, it is necessary to adjust the value of ∂E ∂w⁄ jk. In order to reduce the total error E in the

direction of the gradient of ∂E ∂w⁄ jk, it is assumed that the correction value of neuron j to neuron k is expressed as ∆jk. Let ∆jk be proportional to the error E to the wjk gradient, ie

∆𝑗𝑘= −𝜆𝜕𝑤𝜕𝐸

𝑗𝑘 = −𝜆𝛻 (8)

Where λ is often referred to as the learning rate, reflecting the magnitude of the weight that varies in the direction of the gradient in each cycle of training. If the value of λ is too small, the calculation speed is too slow, which is not conducive to practical application. The setting value is too large, which may cause oscillation of the system. By adjusting λ, the error is quickly converged and the stability of the system is guaranteed. The error correction method between the input layer and the hidden layer is similar to the above, and will not be mentioned here [6-10].

MODEL TESTING

The artificial neural network used in this paper includes an input layer, an excitation layer, a hidden layer, and an output layer, and a dropout is needed to prevent over-fitting. The figure 2 is a data flow diagram of a neural network that reflects the hierarchical structure of the entire model and the conversion relationship of the data.

[image:5.612.255.359.560.659.2]

The figure 3 is the loss function of the model training process. It can be seen from the figure that as the training period changes, the value of the loss function decreases continuously and eventually stabilizes.

(6)
[image:6.612.139.476.88.215.2]

Figure 3. Loss function.

As shown below. Each chart displays the time "slices" of the data, where each slice is a histogram of the tensor at a given step. It is based on the oldest time step principle, with the most recent time step at the top.

Each line on the table in the figure below represents the percentile of the data distribution. For example, the bottom line shows the trend of the minimum value over time, and the middle line shows how the median changes. From top to bottom, each row has the following meaning: [maximum, 93%, 84%, 69%, 50%, 31%, 16%, 7%, minimum]. These percentiles can also be viewed as a normal distribution of standard deviations: [maximum, μ + 1.5 σ, μ + σ, μ + 0.5 σ, μ, μ - 0.5 σ, μ - σ, μ - 1.5 σ, The minimum value] is such that the colored areas read from the inside to the outside have widths [σ, 2σ, 3σ], respectively.

[image:6.612.103.496.455.659.2]
(7)

CONCLUSIONS

In the current parking system, the number of parking spots is too much experience, resulting in unscientific delivery, excessive placement in some places, resulting in waste of resources, and too few places to pay too much to meet demand. In this paper, by analyzing the multi-factors affecting bicycle delivery, the neural network model is used to determine the approximate functional relationship between the statistical data and the delivery forecasting quantity, so as to achieve a more accurate prediction of the number of parking points. Using the neural network model can not only predict the number of bicycles, but also avoid the shortcomings of the regression model, and ensure the accuracy of the prediction. The model calculates the number of delivery to meet the needs of residents while conserving resources, and has certain practicality. Value and practical significance.

ACKNOWLEDGMENTS

This work was supported by the National Natural Science Foundation of China under Grant 61661004, State Key Laboratory of Advanced Optical Communication Systems Networks and Guangxi Science Foundation (2017GXNSFAA198227).

REFERENCES

1. Wen Huiying, Yang Shuzhao. Research on Prediction of Public Bicycle Volume Based on Iterative Regression Method [J]. Journal of Wuhan University of Technology (Transportation Science & Engineering), 2014, 38(02): 245-248.

2. Cao Xuejin, Wang Wei, Ji Yanxi, Zhao Lin-na. Analysis of Influencing Factors of Rail Transit Behavior of Public Bicycle Transfer [J]. Journal of Transportation Engineering and Information, 2015, 13(04): 96-101+119.

3. Jiao Licheng, Yang Shuyuan, Liu Fang,Wang Shigang, Feng Zhiyi. Neural Network for Seventy Years: Retrospect and Prospect [J]. Chinese Journal of Computers, 2016, 39(08): 1697-1716. 4. Nair V., Hinton G.E. Rectified Linear Units Improve Restricted Boltzmann Machines [C].

international conference on machine learning, 2010: 807-814.

5. Arora R., Basu A., Mianjy P., et al. Understanding Deep Neural Networks with Rectified Linear Units [J]. international conference on learning representations, 2018.

6. Wang Jingjing, Wang Jian. Research on an Improved Algorithm of BP Neural Network [J]. Software Guide, 2015, 14(03): 52-53.

7. Glorot X., Bengio Y. Understanding the difficulty of training deep feedforward neural networks [C]. international conference on artificial intelligence and statistics, 2010: 249-256.

8. Schmidhuber J. Deep learning in neural networks [J]. Neural Networks, 2015: 85-117.

9. Klambauer G., Unterthiner T., Mayr A., et al. Self-Normalizing Neural Networks [J]. arXiv preprint arXiv:1706.02515, 2017.

Figure

Figure 2. Model data flow diagram.
Figure 3. Loss function.

References

Related documents

The corona radiata consists of one or more layers of follicular cells that surround the zona pellucida, the polar body, and the secondary oocyte.. The corona radiata is dispersed

Proprietary Schools are referred to as those classified nonpublic, which sell or offer for sale mostly post- secondary instruction which leads to an occupation..

• Follow up with your employer each reporting period to ensure your hours are reported on a regular basis?. • Discuss your progress with

For this solution the Hitachi Command Suite was used to create logical devices (LDEVs) and dynamic pools for the SAP Solution Manager Instance and SAP ERP 6.0 Enhancement Pack 5

4.1 The Select Committee is asked to consider the proposed development of the Customer Service Function, the recommended service delivery option and the investment required8. It

Currently, National Instruments leads the 5G Test & Measurement market, being “responsible for making the hardware and software for testing and measuring … 5G, … carrier

Using text mining of first-opinion electronic medical records from seven veterinary practices around the UK, Kaplan-Meier and Cox proportional hazard modelling, we were able to

National Conference on Technical Vocational Education, Training and Skills Development: A Roadmap for Empowerment (Dec. 2008): Ministry of Human Resource Development, Department