CN107622301B - Method for predicting number of vacant parking positions in parking lot - Google Patents

Abstract

The embodiment of the invention discloses a new prediction method for the number of vacant parking positions in a parking lot, which comprises the steps of preparing data, dividing the data into a training set and a test set; carrying out normalization processing on the data in the training set and the test set; taking the data in the training set as an analysis object, applying a drosophila optimization algorithm to the parameter optimization of the support vector regression model, and selecting the optimal model parameters to establish the support vector regression model with the optimal effect; and substituting the test concentrated data into the support vector regression model with the optimal effect, and predicting the final remaining parking berth number. The implementation of the invention can overcome the defects of over-complex operation, slow operation efficiency, insufficient prediction precision and the like in the prior art.

Description

Method for predicting number of vacant parking positions in parking lot

Technical Field

The invention relates to the technical field of parking guidance berth prediction, in particular to a new prediction method of vacant parking berth number of a parking lot. From the aspect of innovation, the invention creates a new prediction method on the parking guidance berth prediction technology, and compared with the neural network algorithm, the method has the advantages of high stability, less quantity of parameters to be controlled and the like.

Background

With the development of domestic economy, the living standard of people is continuously improved, more and more private cars are used, so that the traffic jam of a city is directly caused, the parking problem gradually becomes a traffic problem, and the situation is more prominent particularly in the urban center of the city. In order to alleviate the problem of parking difficulty, a perfect parking guidance mechanism is needed, which can provide the remaining parking positions of nearby parking lots for the driver, guide the driver to quickly and effectively find available parking spaces, relieve traffic pressure, and reduce exhaust emission.

At present, the instrument is inducted in a lot of parks, for example smart mobile phone APP all is the vacant amount that shows current parking stall in real time, but when reacing the parking area after, the vacant condition in parking stall often can change, has great error with the information that knows earlier. Therefore, people often need to know how many parking spaces exist in a certain future time, so that a journey can be better planned, and the problem of difficulty in parking is solved.

The parking guidance berth prediction is widely researched at home and abroad, and the prediction of the vacant parking berth number of the parking lot belongs to the time series prediction problem. Current research is broadly divided into two categories: one is single step prediction, namely, the residual vacant parking space number of a single time node in the future is predicted by using historical data of the parking vacant parking spaces, for example, the vacant parking space number at t +10 is predicted by using t-40, t-30, t-20, t-10 and t; the second type is multi-step prediction, namely, firstly, historical data of the vacant parking spaces are used for predicting a single point in the future, then the predicted point is added into the historical data for predicting the next point, parking space information in a future period of time is predicted in an iterative mode, for example, the number of the vacant parking spaces in t +10 time is predicted by using t-40, t-30, t-20, t-10 and t, then the number of the vacant parking spaces in t +20 time is predicted by using t-30, t-20, t-10, t and t +10, and then the parking space information in the future period of time is continuously predicted in an iterative mode.

In the single-step prediction method, M.Caliskan, A.Barthels, B.Scheuermann, M.Mauve et al used a queuing theory-based continuous time uniform Markov model to predict the short-time remaining parking berth number and achieve good prediction effect. A prediction method combining phase space reconstruction and Elman neural network is introduced in Chenqun, Yankefei, Wanrentao, Moyikui and the like in 2007, and the prediction method has better prediction accuracy and effectiveness through prediction examples. The quaternary bending arm, Wangwei and Dengder (2007) provide an effective parking vacant parking berth prediction method based on a weighted Markov chain of wavelet analysis. The time sequence is decomposed and reconstructed through the wavelet, and the prediction precision is improved. According to the characteristics of effective parking position change, Homowave, Han Ying and Yao (2012), a chaotic time series method is adopted to carry out phase space reconstruction on historical data of the parking position, a BP neural network model is established to predict the change trend of the effective parking position, and related experiments prove that the method is small in relative error and high in prediction precision. Andreas klapencecker, hyuneung Lee, Jennifer l.welch et al (2014) utilize queuing theory and continuous-time uniform markov models and simplify the calculation of transition probabilities in vehicle navigation systems. Zhao Wu Chen, Zhang Yu Ru (2015) adopts BP neural network momentum method and adjusts the method that the learning rate combines to improve it, verified the BP neural network through the emulation to the validity that the vacant berth number of parking area predicts. Chenhaipeng, Xiaohang et al (2017) predict short space-time residual parking positions by a method of combining wavelet transformation and an Extreme Learning Machine (ELM). Firstly, performing wavelet decomposition and reconstruction on an effective parking position time sequence through a wavelet function; then predicting each time sequence obtained after decomposition by using ELM; and finally, synthesizing the prediction results of the neural networks to obtain the final prediction result.

However, currently, the focus is mainly on short-time single-step prediction, and the multi-step prediction involves less. For example, Liu, Shixu, Guan, Hongzhi et al (2010) develops a weighted first-order local area method to predict unoccupied parking spaces based on a chaotic time series prediction method of historical data, and performs multi-step prediction by using an iterative method, and the result shows that the prediction model is accurate and has great practical value. JI Yan-jie (Jie) 1, TANG Dou-nan (Tang Tong nan) et al (2014) propose a novel multi-step prediction model named as WNN-LE method, and improve the accuracy and stability of multi-step prediction by combining a wavelet neural network model and a maximum Lyapunov exponent method. Jia Jian super et al (2013) utilizes a maximum Lyapunov exponent method to judge the chaos characteristic of a traffic flow time sequence, carries out phase space reconstruction on the traffic flow time sequence, and designs a multi-step traffic flow prediction algorithm based on a chaos theory by combining a weighted first-order local area method on the basis. Fengquan Yu et al (2015) select an ARIMA model to predict unoccupied parking spaces, establish a residual parking space prediction model according to a general flow of the ARIMA model, further combine actual data, test the prediction accuracy, compare the prediction accuracy with the neural network prediction effect, and verify the effectiveness and applicability of the ARIMA model in predicting residual parking spaces.

However, the inventors have found that the conventional method for predicting the number of vacant parking spaces in a parking lot has disadvantages in that the calculation is complicated, the operation efficiency is slow, and the prediction accuracy is not high enough.

Disclosure of Invention

The embodiment of the invention aims to provide a method for predicting the number of vacant parking positions in a parking lot, which can overcome the defects of over-complex operation, low operation efficiency, low prediction precision and the like in the prior art.

In order to solve the above technical problem, an embodiment of the present invention provides a method for predicting a number of vacant parking spaces in a parking lot, including:

s1, preparing data, wherein the data comprises a training set formed by the residual parking positions of the known parking lots and a testing set formed by the residual parking positions of the estimated parking lots;

s2, passing the formula x by adopting a maximum and minimum method_k＝(x_k-x_min)/(x_max-x_min) Normalizing the data in the training set and the test set;

s3, setting the population scale, the maximum iteration times, the initial position range of the fruit fly population and the single flight range of the fruit flies; wherein the initial position range of the drosophila colony is given by corresponding two-dimensional coordinates (X _ axis, Y _ axis), specifically

S4, passing formula

Calculating the random flight direction and distance of fruit fly individuals in the training set for searching food by using smell, and calculating the random flight direction and distance by using a formula

Calculating the distance between the individual drosophila in the training set and the origin and a taste concentration judgment value, and further substituting the taste concentration judgment value into a preset taste concentration judgment function to obtain the taste concentration value of the individual drosophila in the training set; wherein the taste concentration determination value is the reciprocal of the distance from the individual drosophila to the origin; the preset taste concentration judgment function is Smell_iBestc and bestg are parameters, p _ train is a taste concentration judgment value calculated by the training set, and t _ train is a taste concentration judgment value calculated by the training set;

s5, substituting the parameter bestc as a penalty factor and bestg as a kernel function parameter into a preset support vector regression model, and determining the corresponding parameter bestc and bestg when the mean square error obtained by the preset support vector regression model is minimum;

s6, finding out individuals with the optimal taste in the drosophila population according to the determined parameters bestc and bestg, and recording the optimal taste concentration value of the drosophila individuals;

s7, keeping the optimal taste concentration value and the corresponding position information of the fruit flies, and enabling other fruit flies in the group to fly to the position by using vision to form a new bunching relation;

s8, repeating the steps S4 to S7, and performing iterative optimization until the maximum iteration times are reached to obtain final parameters bestc and bestg;

and S9, substituting the estimated remaining parking positions of the parking lot in the test set and the final parameters bestc and bestg into the preset support vector regression model to obtain the final predicted remaining parking positions.

The embodiment of the invention has the following beneficial effects:

the method takes the training centralized data as an analysis object, applies the drosophila optimization algorithm to the parameter optimization of the support vector regression model, selects the optimal model parameters to establish the support vector regression model with the optimal effect to predict the final residual parking berth number, not only accelerates the convergence speed of the algorithm, but also improves the prediction precision.

Because different data sets adopted by different papers are difficult to compare, the comparison of prediction precision is carried out on an article (BP neural network is used for predicting and researching vacant parking spaces in parking lots, namely Zhao, Zhang Yu Ru) which adopts the same data set as the article, the higher EC value is, the higher prediction precision is indicated, and the value range is (0, 1).

TABLE 1 comparison of error between predicted and actual results

To show that the prediction accuracy is higher than other prediction methods, the method is compared with the method accuracy using different data articles. Table 2 compares the method with different data set articles (effective parking berth prediction based on chaos and BP neural networks-flood wave and parking guidance berth prediction based on BP neural networks-high wide bank and intelligent parking effective berth prediction research based on markov prediction model-korean second peak).

TABLE 2 comparison of error between predicted and actual results

TABLE 3 Drosophila optimized support vector regression model 5 times of experimental results

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is within the scope of the present invention for those skilled in the art to obtain other drawings based on the drawings without inventive exercise.

Fig. 1 is a flowchart of a method for predicting a number of vacant parking positions in a parking lot according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

As shown in fig. 1, a method for predicting a number of vacant parking spaces in a parking lot according to an embodiment of the present invention includes:

step S1, preparing data, wherein the data comprises a training set formed by the residual parking positions of the known parking lot and a testing set formed by the residual parking positions of the estimated parking lot;

the specific process is that the parking data is obtained by recording the remaining number of parking lots at regular intervals (for example, 10 minutes), for example: the parking lot parking space number is determined by the following steps of 3:10 min-192, 3:20 min-195, 3:30 min-194, 3:40 min-193, 3:50 min-193 and 4:00 min-195.

The residual parking number of the sixth time point in the test set is predicted by forming training set data at the first 5 time points, and then the error comparison is carried out on the residual parking number and the actual parking number.

S2, passing the formula x by adopting a maximum and minimum method_k＝(x_k-x_min)/(x_max-x_min) Normalizing the data in the training set and the test set;

in particular toThe process is that because the experimental data is difficult to collect, the experiment adopts the vacant parking position data of a certain parking lot for two continuous days in a certain time period to carry out the experiment, the more the data quantity is, the more the model performance can be reflected, the collected data is divided into a training set and a testing set, then the data is normalized, and the normalization formula x_k＝(x_k-x_min)/(x_max-x_min) The maximum-minimum method is used.

S3, setting the population scale, the maximum iteration times, the initial position range of the fruit fly population and the single flight range of the fruit flies; wherein the initial position range of the drosophila colony is given by corresponding two-dimensional coordinates (X _ axis, Y _ axis), specifically

The specific process comprises the steps of setting relevant parameter values such as population size (popsize), maximum iteration number (maxgen), fruit fly population position range and single flying range of fruit flies. The positional information of each individual in the drosophila population is given by the corresponding (X, Y) two-dimensional coordinates. Its initial position is given by the following formula definition:

s4, passing formula

Calculating the random flight direction and distance of fruit fly individuals in the training set for searching food by using smell, and calculating the random flight direction and distance by using a formula

Calculating the distance between the individual drosophila in the training set and the origin and a taste concentration judgment value, and further substituting the taste concentration judgment value into a preset taste concentration judgment function to obtain the taste concentration value of the individual drosophila in the training set; wherein the taste concentration determination value is the reciprocal of the distance from the individual drosophila to the origin; the preset taste concentration determination function is Smell_iBestc and bestg are parameters, p _ train is a taste concentration judgment value calculated by the training set, and t _ train is a taste concentration judgment value calculated by the training set;

the specific process is that when each fruit fly in the population searches by using the smell, each fruit fly is endowed with a random flight direction and distance. The new position of individual drosophila i is given by:

since the origin of the food (reference parameter) taste is unknown, the distance Dist of the individual drosophila from the origin is first calculated using the following formula_i。

Then, the taste concentration judgment value S is calculated by the following formula_i；

The taste concentration value Smell of each individual Drosophila in the current population was calculated by the following formula_i；

Smell_i＝fitness(bestc,bestg,p_train,t_train) (5)

The fitness represents a taste concentration judgment function, and is an objective function or a fitness function when the FOA is used for solving an optimization problem; bestc and bestg are parameters, p _ train is a taste concentration determination value calculated for the training set, and t _ train is a taste concentration determination value calculated for the training set.

S5, substituting the parameter bestc as a penalty factor and bestg as a kernel function parameter into a preset support vector regression model, and determining the corresponding parameter bestc and bestg when the mean square error obtained by the preset support vector regression model is minimum;

the specific process is to bring the parameters bestc and bestg into a Support Vector Regression (SVR) model to judge which set of c and g can minimize the Mean Square Error (MSE) of the model. The basic idea of SVR is to map a high-dimensional feature space F through a non-linearity and perform a linear regression in the space F, where the regression function is:

f(x)＝(w·φ(x))+b (6)

where w is the weight vector and b is the bias term.

By using an insensitive loss function with sparsity, the constraint optimization problem is expressed as:

bestc is a penalty factor and is used for balancing the flatness of the regression model and the number of sample points with deviation larger than the deviation. Xi_iAnd

is a relaxation factor. The formula (7) is a typical convex quadratic programming problem, a lagrange function is introduced to convert the convex quadratic programming problem into a dual optimization problem, and the dual optimization problem is solved according to a KKT condition to obtain:

bringing equation (8) into equation (6) yields a regression function as:

wherein, k (x)_i,x)＝φ(x_i)φ(x_j) Is a kernel function, which isSymmetric positive real functions, while satisfying the Mercer theorem. The present algorithm here employs the more commonly used radial basis kernel function.

Wherein bestg >0 is the bandwidth of the gaussian kernel.

S6, finding out individuals with the optimal taste in the drosophila population according to the determined parameters bestc and bestg, and recording the optimal taste concentration value of the drosophila individuals;

the specific process is to select the value with the best taste concentration in the current population (Smell)_i) The taste concentration and the corresponding positions (bestc and bestg) were recorded.

S7, keeping the optimal taste concentration value and the corresponding position information of the fruit flies, and enabling other fruit flies in the group to fly to the position by using vision to form a new bunching relation;

the specific process is that the optimal taste concentration value and the corresponding fruit fly position information are maintained, and other fruit flies in the group fly to the position by using vision, namely

Smellbest＝bestSmell

X_axis＝X(bestIndex)

Y_axis＝Y(bestIndex)

S8, repeating the steps S4 to S7, and performing iterative optimization until the maximum iteration times are reached to obtain final parameters bestc and bestg;

the specific process is that the optimal model parameters are selected to establish the optimal support vector regression model until the maximum iteration times.

And S9, substituting the estimated remaining parking positions of the parking lot in the test set and the final parameters bestc and bestg into the preset support vector regression model to obtain the final predicted remaining parking positions.

The specific process is that according to the support vector regression model with the optimal effect, the final remaining parking berth number is predicted by predicting the remaining berth number of the parking lot.

The embodiment of the invention has the following beneficial effects:

the method takes the training centralized data as an analysis object, applies the drosophila optimization algorithm to the parameter optimization of the support vector regression model, selects the optimal model parameters to establish the support vector regression model with the optimal effect to predict the final residual parking berth number, not only accelerates the convergence speed of the algorithm, but also improves the prediction precision.

It will be understood by those skilled in the art that all or part of the steps in the method for implementing the above embodiments may be implemented by relevant hardware instructed by a program, and the program may be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disk, optical disk, etc.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (1)

1. A method for predicting the number of vacant parking positions in a parking lot is characterized by comprising the following steps:

s1, preparing data, wherein the data comprises a training set formed by the residual parking positions of the known parking lots and a testing set formed by the residual parking positions of the estimated parking lots;

s2, passing the formula x by adopting a maximum and minimum method_k＝(x_k-x_min)/(x_max-x_min) Normalizing the data in the training set and the test set;

s3, setting the population scale, the maximum iteration times, the initial position range of the fruit fly population and the single flight range of the fruit flies; wherein the initial position range of the drosophila colony is given by corresponding two-dimensional coordinates (X _ axis, Y _ axis), specifically

S4, passing formula

Calculating the random flight direction and distance of fruit fly individuals in the training set for searching food by using smell, and calculating the random flight direction and distance by using a formula

Calculating the distance between the individual drosophila in the training set and the origin and a taste concentration judgment value, and further substituting the taste concentration judgment value into a preset taste concentration judgment function to obtain the taste concentration value of the individual drosophila in the training set; wherein the taste concentration determination value is the reciprocal of the distance from the individual drosophila to the origin; the preset taste concentration judgment function is Smell_iBestc and bestg are parameters, p _ train is a taste concentration judgment value calculated by the training set, and t _ train is a taste concentration judgment value calculated by the training set;

s5, substituting the parameter bestc as a penalty factor and bestg as a kernel function parameter into a preset support vector regression model, and determining the corresponding parameter bestc and bestg when the mean square error obtained by the preset support vector regression model is minimum;

the specific process is that parameters bestc and bestg are brought into a support vector regression model to judge which group of c and g can minimize the mean square error of the model, the basic idea of SVR is to map to a high-dimensional feature space F through a non-linearity, and to perform linear regression in the space F, and the regression function is as follows:

f(x)＝(w·φ(x))+b (6)

wherein w is a weight vector and b is an offset term;

by using an insensitive loss function with sparsity, the constraint optimization problem is expressed as:

bestc is a punishment factor used for balancing the flatness of the regression model and the number of sample points with deviation larger than xi_iAnd

for the relaxation factor, the formula (7) is a typical convex quadratic programming problem, a lagrange function is introduced to convert the convex quadratic programming problem into a dual optimization problem, and the dual optimization problem is solved according to a KKT condition to obtain:

bringing equation (8) into equation (6) yields a regression function as:

wherein, k (x)_i,x)＝φ(x_i)φ(x_j) Is a kernel function, which is a symmetric positive real function, and satisfies the Mercer theorem, here a radial basis kernel function is used,

wherein bestg >0 is the bandwidth of the Gaussian kernel;

s6, finding out individuals with the optimal taste in the drosophila population according to the determined parameters bestc and bestg, and recording the optimal taste concentration value of the drosophila individuals;

s7, keeping the optimal taste concentration value and the corresponding position information of the fruit flies, and enabling other fruit flies in the group to fly to the position by using vision to form a new bunching relation;

the specific process is that the optimal taste concentration value and the corresponding fruit fly position information are maintained, and other fruit flies in the group fly to the position by using vision, namely

Smellbest＝bestSmell

X_axis＝X(bestIndex)

Y_axis＝Y(bestIndex)

S8, repeating the steps S4 to S7, and performing iterative optimization until the maximum iteration times are reached to obtain final parameters bestc and bestg;

and S9, substituting the estimated remaining parking positions of the parking lot in the test set and the final parameters bestc and bestg into the preset support vector regression model to obtain the final predicted remaining parking positions.

Images