WO2024065954A1 - 停车场泊位占用率短时域预测方法、系统、设备及终端 - Google Patents

停车场泊位占用率短时域预测方法、系统、设备及终端 Download PDF

Info

Publication number
WO2024065954A1
WO2024065954A1 PCT/CN2022/129957 CN2022129957W WO2024065954A1 WO 2024065954 A1 WO2024065954 A1 WO 2024065954A1 CN 2022129957 W CN2022129957 W CN 2022129957W WO 2024065954 A1 WO2024065954 A1 WO 2024065954A1
Authority
WO
WIPO (PCT)
Prior art keywords
time series
series data
occupancy rate
parking lot
parking
Prior art date
Application number
PCT/CN2022/129957
Other languages
English (en)
French (fr)
Inventor
晏鹏宇
谢皓宇
蔡小强
Original Assignee
电子科技大学长三角研究院(湖州)
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 电子科技大学长三角研究院(湖州) filed Critical 电子科技大学长三角研究院(湖州)
Publication of WO2024065954A1 publication Critical patent/WO2024065954A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/02—Reservations, e.g. for tickets, services or events
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/40—Business processes related to the transportation industry

Definitions

  • the present invention belongs to the technical field of data analysis, and in particular relates to a method, system, device and terminal for short-term prediction of parking lot occupancy rate.
  • the parking lot occupancy rate prediction methods at home and abroad mainly use the univariate time series data of the parking lot historical parking lot occupancy rate to predict the parking lot occupancy rate in the future in days or weeks.
  • some methods consider the prediction of parking lot occupancy rate in short time domains (such as 5 minutes to 180 minutes).
  • the parking lot reservation results in the short time domain will provide real-time data on the use of parking resources for static transportation services such as parking reservations in the future smart city construction.
  • the parking lot occupancy rate in the short time domain is affected by many random factors, it is difficult to fully capture the law of parking lot occupancy rate in limited data. Therefore, this type of prediction method is currently ineffective and cannot accurately predict the changes in parking lot occupancy rate in a short period of time.
  • these prediction methods treat the parking lot historical occupancy rate data, which is a time series data, as a piece of data and split it into a training set and a test set in chronological order.
  • the data of the first part of the time is used as the training part of the model, that is, the model learns only the rules of the data of the first part of the time, and the rules of the data of the second part of the time cannot be effectively learned. Therefore, when the rules of the data of the second part are greatly different from those of the first part, the prediction effect of the model will be lacking.
  • the short-term prediction method of parking lot occupancy rate based on the time series data matrix proposed by the present invention effectively solves this problem.
  • one of the problems and defects of the existing technology is: since the berth occupancy rate in the short-term domain is affected by a variety of random factors, the changing patterns of the previous and subsequent data are often different, and the data processing method of the existing technology can only enable the prediction method to learn the berth occupancy rate pattern in the first half. Therefore, the current short-term berth prediction method has poor prediction effect on the second half of the data, and cannot accurately predict the changes in berth occupancy rate in a shorter time.
  • the present invention provides a method, system, device and terminal for short-term prediction of parking space occupancy rate, and more particularly, relates to a method, system, device and terminal for short-term prediction of parking space occupancy rate based on a time series data matrix.
  • the present invention is implemented as follows: a method for short-term prediction of parking space occupancy rate in a parking lot, the method comprising: utilizing vehicle entry and exit records recorded in the background of a parking fee management system, obtaining historical parking space occupancy data through data sorting and statistics, using the time series matrix construction method proposed in the present invention to construct a time series data matrix as data to be used in subsequent prediction methods, and randomly selecting data in the matrix in combination with a gradient ascent decision tree model to train and test the model, thereby improving the accuracy of short-term prediction of parking space occupancy rate.
  • the parking lot occupancy rate short-term prediction method comprises the following steps:
  • Step 1 Using the vehicle entry and exit records recorded in the parking fee management system background, through data sorting and statistics, the historical parking space occupancy rate data is obtained.
  • the data acquisition is simple and does not involve privacy data, which provides a data basis for the subsequent construction of the time series data matrix;
  • Step 2 Use the time series data matrix construction method to split the time series data into time series data subsets of the same length and construct a time series data matrix.
  • Each row of data in the matrix is a piece of data in an adjacent time period that can be used for subsequent prediction model training and testing, ensuring that the model can learn the law of berth occupancy rate changes when randomly selecting data in the future;
  • Step 3 randomly select data from the time series data matrix, combine it with the gradient ascent decision tree model, and train the model. Randomly select data so that the model can learn the changing rules of the berth occupancy rate within the entire berth occupancy rate recording period.
  • the gradient ascent decision tree model itself has strong generalization performance, which improves the prediction accuracy of the berth occupancy rate.
  • Step 4 Use the trained gradient ascent decision tree model to predict the parking lot occupancy rate. This is the final output of this method and can produce more accurate prediction results than a model trained with a time series data.
  • the parking space occupancy rate time series data in step one refers to parking space occupancy rate data of equal time intervals constructed in chronological order; the parking space occupancy rate refers to the ratio of the number of parking spaces used in a parking lot at a certain moment to the total number of parking spaces, and the time interval can be of any length.
  • the time interval is preferably selected to be 5 to 180 minutes.
  • time series data matrix construction method in step 2 is used to split a berth occupancy time series data into time series data subsets of the same size to construct a time series data matrix.
  • the time series data matrix construction method specifically includes:
  • the length is T, where O t represents the parking space occupancy rate value of the parking lot at time t;
  • the gradient ascent decision tree model in step three refers to serially generating a series of CART regression trees, each tree continuously fits the residual after learning the previous tree, and gradually learns the model of the true value, and its training data is randomly selected data from the time series data matrix.
  • the data in the time series data matrix is divided into a training set and a test set in an arbitrary ratio; wherein the ratio of the training set to the test set is preferably 7:3.
  • Another object of the present invention is to provide a parking lot parking space occupancy rate short-term domain prediction system using the parking lot parking space occupancy rate short-term domain prediction method, the parking lot parking space occupancy rate short-term domain prediction system comprising:
  • a time series data construction module is used to construct parking space occupancy rate time series data based on the historical parking space occupancy rate data of the parking lot;
  • a time series data matrix construction module is used to split the time series data into time series data subsets of the same length through a time series data matrix construction method to construct a time series data matrix;
  • the model training module is used to randomly select data from the time series data matrix and train the model by combining the gradient ascent decision tree model;
  • the parking lot occupancy prediction module is used to predict the parking lot occupancy rate using the trained gradient ascent decision tree model.
  • Another object of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the parking lot space occupancy short-term domain prediction method.
  • Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the parking lot space occupancy rate short-term domain prediction method.
  • Another object of the present invention is to provide an information data processing terminal, which is used to implement the parking lot space occupancy rate short-term domain prediction system.
  • the short-term domain prediction method for parking lot occupancy mainly utilizes the historical parking occupancy data recorded in the parking lot background, and its core lies in using the original data to construct a time series data matrix.
  • the historical parking occupancy data of the parking lot is obtained to construct an occupancy time series data; then, the time series data matrix construction method proposed by the present invention is used to construct the original data into a time series data matrix, and each data in the matrix is a subset of the original time series data; secondly, the data in the time series data matrix is randomly selected, and the model is trained in combination with the gradient ascent decision tree model; finally, the trained gradient ascent decision tree model is used to predict the parking lot occupancy.
  • the present invention expands a time series data used in the conventional parking occupancy prediction into a time series data matrix.
  • the time series data is randomly selected for the parking occupancy prediction model to be trained, so that the model can better learn the law of parking occupancy change and improve the prediction accuracy of parking occupancy.
  • the present invention provides a short-term prediction method for parking lot parking space occupancy rate based on a time series data matrix.
  • the vehicle entry and exit records recorded in the background of the parking lot fee management system are used to obtain historical parking space occupancy rate data through data sorting and statistics.
  • the time series matrix construction method proposed by the present invention is used to construct a time series data matrix as data to be used in subsequent prediction methods.
  • Each row of data in the matrix is a piece of data in an adjacent time period that can be used for subsequent prediction model training and testing, ensuring that the model can learn the law of parking space occupancy rate changes when randomly selecting data in the subsequent period.
  • the data in the matrix is randomly selected and combined with the gradient ascent decision tree model to train and test the model. The randomly selected data enables the model to learn the parking space occupancy rate change law within the entire parking space occupancy rate recording time period, thereby improving the accuracy of short-term prediction of parking lot parking space occupancy rate.
  • the expected benefits and commercial value of the technical solution of the present invention after transformation are as follows: At present, major cities have begun to promote the construction of city-level smart parking systems, which integrate query, reservation, navigation and charging. The method is embedded in the smart parking system to provide a more accurate prediction of parking space occupancy for the smart parking reservation system.
  • the beneficial effects are as follows: first, more accurate prediction of parking space occupancy. When users query and reserve parking spaces, they can better understand the congestion of parking lots when they arrive near their destinations.
  • more accurate prediction of parking space occupancy can help the system better recommend parking lots and make parking lot pricing decisions for users, and can give priority to recommending vacant parking lots, effectively adjust the balance of parking space occupancy in the region, and reduce traffic congestion and environmental pollution caused by crowded parking lots.
  • FIG1 is a flow chart of a method for predicting parking space occupancy in a short time domain according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a method for constructing a time series data matrix provided by an embodiment of the present invention to construct a time series data matrix with a length of T into a time series data matrix with N columns;
  • Figure 3 is a comparison chart of the parking space occupancy prediction results on the test set in a parking lot provided by an embodiment of the present invention, after training using a time series data sequence to select time series data (conventional method) and a time series data matrix to randomly select time series data (method proposed by the present invention) as the training set of the model.
  • the present invention provides a method, system, device and terminal for short-term prediction of parking space occupancy rate.
  • the present invention is described in detail below with reference to the accompanying drawings.
  • the short-term prediction method for parking lot occupancy rate includes the following steps:
  • the parking space occupancy time series data provided in the embodiment of the present invention refers to parking space occupancy data of equal time intervals constructed in chronological order.
  • the parking space occupancy rate refers to the ratio of the number of parking spaces used in a parking lot at a certain moment to the total number of parking spaces.
  • the time interval can be of any length; preferably, the time interval is selected from 5 to 180 minutes.
  • the method for constructing a time series data matrix provided in an embodiment of the present invention is used to split a berth occupancy rate time series data into time series data subsets of the same size to construct a time series data matrix.
  • the time series data subset refers to the parking space occupancy rate time series data with a length less than T, and the data in the time series data subset are the parking space occupancy rates at adjacent times, such as (O 1 , O 2 , O 3 ) is a subset of the above time series data O.
  • the gradient ascent decision tree model provided in the embodiment of the present invention refers to a model that generates a series of CART regression trees in series, where each tree continuously fits the residual after learning the previous tree, and gradually learns the true value, and its training data is data randomly selected from the time series data matrix.
  • the data in the time series data matrix is divided into a training set and a test set in an arbitrary ratio, such as 7:3.
  • a time series data construction module is used to construct parking space occupancy rate time series data based on the historical parking space occupancy rate data of the parking lot;
  • a time series data matrix construction module is used to split the time series data into time series data subsets of the same length through a time series data matrix construction method to construct a time series data matrix;
  • the model training module is used to randomly select data from the time series data matrix and train the model by combining the gradient ascent decision tree model;
  • the parking lot occupancy prediction module is used to predict the parking lot occupancy rate using the trained gradient ascent decision tree model.
  • the parking lot parking space occupancy rate short-term domain prediction method based on the time series data matrix uses the historical parking space occupancy rate data recorded in the parking lot backend to construct a time series data matrix, and randomly selects data in the matrix and combines the gradient ascent decision tree model to train and predict the model.
  • the time series data matrix construction method provided by the embodiment of the present invention is:
  • the corresponding time series data subset is There are 3 subsets in total, and they are concatenated into a matrix by row.
  • the embodiment of the present invention provides 782362 real data values of parking in a hospital in a provincial capital city during the period from August 1, 2019 to August 21, 2020, using a time series data sequence to select time series data (conventional method) and a time series data matrix to randomly select time series data (method proposed in the embodiment of the present invention) as the training set of the model, and the berth occupancy prediction result on the test set;
  • the embodiment of the present invention respectively adopts three commonly used berth occupancy prediction models: linear regression, gradient ascent decision tree, neural network, and the evaluation index is the mean absolute error, which is expressed as follows:
  • M is the number of time series data subsets in the test set, is the predicted value of the model at time t, and y t is the true value of the parking lot at time t.
  • FIG3 a comparison diagram of berth occupancy prediction results of three commonly used berth occupancy prediction models is shown in an embodiment of the present invention. It can be seen that the use of the time series data matrix of the present invention to select a training set can reduce the mean absolute error of the prediction, that is, it can effectively improve the prediction accuracy.
  • the traditional data processing method that is, the training set and the test set are separated in chronological order, and the data of the training set (the data of the first part of the data recording period) is distributed according to the random variable where ⁇ 1 is the mean of X 1 , is the variance of X1 , and the data of the test set (the data of the latter part of the data recording period) is distributed according to the random variable
  • the data distribution learned by the model is only X1 .
  • a time series data is constructed into a time series data matrix, and after randomly selecting data to provide training for the model, the data distribution learned by the model is the weighted average of X1 and X2 .
  • Case 1 Random variables X1 and X2 obey the same or similar distribution, that is, the data change rules in the period before and after the data recording time period are the same or similar. Assume that through the traditional data processing method and the time series data matrix proposed by the present invention, the model training results have the same or similar distribution as X1 , and the prediction results of the two on the test set are also the same or similar. In fact, the effect of our proposed method is not worse than that of the traditional method.
  • Case 2 The distributions of random variables X1 and X2 are very different. We can assume that X1 and X2 are independent of each other. Assuming that the ratio of the training set to the test set is a:b, the data distribution of the model learning in the traditional method obeys the random variable The mean absolute error with the test set is: The data distribution of the model learning in the method proposed by the present invention obeys the random variable It is similar to the test set (whose data distribution also follows ) is: That is, when the data in the training set and the test set only obey one distribution, ideally, the method we proposed will reduce the prediction error to 0, which is better than the traditional method.
  • the prediction error is generally not 0, because the proportion of random variables with the same distribution in the training set and the test set is not necessarily the same, and these data often do not obey only one distribution, that is, the data distribution learned by the model is only the weighted average of many random variables.
  • the method we proposed can effectively reduce the error caused by the different distribution of random variables obeyed by the training set and the test set, making the model's prediction effect better.
  • the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware.
  • the hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware.
  • an appropriate instruction execution system such as a microprocessor or dedicated design hardware.
  • the above-mentioned devices and methods can be implemented using computer executable instructions and/or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier.
  • a processor control code such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier.
  • a carrier medium such as a disk, CD or DVD-ROM
  • a programmable memory such as a read
  • the device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Operations Research (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • Quality & Reliability (AREA)
  • Computational Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Game Theory and Decision Science (AREA)
  • Algebra (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种停车场泊位占用率短时域预测方法、系统、设备及终端，属于数据分析技术领域。该方法包括：基于停车场的历史泊位占用率数据，构建泊位占用率时序数据（S101）；通过时序数据矩阵构建方法，将时序数据拆分为长度相同的时序数据子集，构建时序数据矩阵（S102）；随机选取时序数据矩阵中的数据，结合梯度上升决策树模型，对模型进行训练（S103）；使用训练后的梯度上升决策树模型对停车场的占用率进行预测（S104）。将常规泊位占用率预测中使用的时序数据，扩充为一个时序数据矩阵，在时序数据矩阵的基础上，随机选取矩阵中的数据结合梯度上升决策树模型对模型进行训练和预测，从而使模型能更好地学习到泊位占用率变化的规律，提高停车场泊位占用率短时域预测的准确度。

Description

停车场泊位占用率短时域预测方法、系统、设备及终端 技术领域
本发明属于数据分析技术领域，尤其涉及一种停车场泊位占用率短时域预测方法、系统、设备及终端。
背景技术
目前，国内外停车场泊位占用率预测方法主要使用停车场历史泊位占用率这一单变量时间序列数据对未来较长时域内以天或者周为时间单位的停车场泊位占用率进行预测。近年来随着智慧停车技术的发展，部分方法考虑短时域(如5分钟～180分钟)内泊位占用率预测。短时域的泊位预约结果将为未来智慧城市建设中预约停车等静态交通服务提供泊位资源使用的实时数据。但由于短时域的泊位占用率受到多种随机因素的影响，在有限的数据中很难完整捕捉到泊位占用率的规律，因此目前这类预测方法效果欠佳，无法较为准确的预测泊位占用率在较短时间内的变化。其原因之一在于这些预测方法对于停车场历史占用率数据这一时序数据的处理方式都是将其作为一条数据按照时间先后顺序进行训练集和测试集的拆分，前部分时间的数据作为模型的训练部分，也就是说模型学习到的只是前部分时间数据的规律，且后部分时间的数据规律并不能进行有效学习，因此当后部分的数据规律与前部分差异较大时，会导致模型的预测效果有所欠缺。而本发明提出的基于时序数据矩阵的停车场泊位占用率短时域预测方法有效地解决了这一问题。
通过上述分析，现有技术存在的问题及缺陷之一为：由于短时域的泊位占用率受到多种随机因素的影响，前后数据的变化规律往往不同，而现有技术的数据处理方法只能使得预测方法学习到前半段的泊位占用率规律，故目前短时域的泊位预测方法在后半段数据上的预测效果欠佳，无法较为准确的预测泊位占用率在较短时间内的变化。
发明内容
针对现有技术存在的问题，本发明提供了一种停车场泊位占用率短时域预测方法、系统、设备及终端，尤其涉及一种基于时序数据矩阵的停车场泊位占用率短时域预测方法、系统、设备及终端。
本发明是这样实现的，一种停车场泊位占用率短时域预测方法，所述停车场泊位占用率短时域预测方法包括：利用停车场收费管理系统后台记录的车辆出入记录，通过数据整理与统计，得到历史泊位占用率数据，使用本发明提出的时序矩阵构建方法，构建出时序数据矩阵作为后续预测方法的将使用到的数据，并随机选取矩阵中的数据结合梯度上升决策树模型对模型进行训练和测试，提高停车场泊位占用率短时域预测的准确度。
进一步，所述停车场泊位占用率短时域预测方法包括以下步骤：
步骤一，利用停车场收费管理系统后台记录的车辆出入记录，通过数据整理与统计，得到历史泊位占用率数据，数据获取简单，不涉及隐私数据，为后续时序数据矩阵的构建提供了数据基础；
步骤二，使用时序数据矩阵构建方法，将时序数据拆分为长度相同的时序数据子集，构建时序数据矩阵，矩阵中的每一行数据都是一条可供后续预测模型训练与测试的一条相邻时间段内的数据，保证后续随机选取数据时模型能学习到泊位占用率变化的规律；
步骤三，随机选取时序数据矩阵中的数据，结合梯度上升决策树模型，对模型进行训练，随机选取数据使得模型能学习到整个泊位占用率记录时间段内的泊位占用率变化规律，同时梯度上升决策树模型本身具有较强的泛化性能，使得泊位占用率的预测准确度上升；
步骤四，使用训练后的梯度上升决策树模型对停车场的占用率进行预测，这是本方法最终输出的部分，相比用一条时序数据进行训练的模型得出更加准确的预测结果。
进一步，所述步骤一中的泊位占用率时序数据，是指按时间顺序构建的等时间间隔的停车场泊位占用率数据；泊位占用率是指某时刻停车场已使用的车位数与总车位数的比值，时间间隔取任意时长。
进一步，构建泊位占用率时序数据时，所述时间间隔优先选取5～180分钟。
进一步，所述步骤二中的时序数据矩阵构建方法，用于根据一条泊位占用率时序数据，将其拆分为大小相同的时序数据子集，构建一个时序数据矩阵。
所述时序数据矩阵构建方法具体包括：
当某停车场的在某天运营时段内泊位占用率时序数据为O＝(O 1,O 2,…,O t,…,O T)，长度为T，其中O t表示t时刻停车场的泊位占用率值；所述时序数据子集是指长度小于T的泊位占用率时序数据，且时序数据子集中的数据为相邻时刻的泊位占用率，如(O 1,O 2,O 3)为前述时序数据O的子集；建立列数为N的时序数据矩阵，对应的时序数据子集为
Figure PCTCN2022129957-appb-000001
Figure PCTCN2022129957-appb-000002
共有M＝T-N+1个子集，并将其按行拼接为时序数据矩阵；当时序数据矩阵的列数为N，实际意义为使用前N-1个时刻的泊位占用率数据，作为模型输入，预测第N个时刻的泊位占用率，作为模型输出。
进一步，所述步骤三中的梯度上升决策树模型，是指串行生成一系列CART回归树，每一颗树不断拟合上一棵树学习后的残差，逐渐学习到真实值的模型，其训练数据为所述时序数据矩阵中随机选取的数据。
在梯度上升决策树模型的训练与测试中，将时序数据矩阵中的数据按任意比例分为训练集和测试集；其中，所述训练集和测试集的比例优选为7：3。
本发明的另一目的在于提供一种应用所述的停车场泊位占用率短时域预测方法的停车场泊位占用率短时域预测系统，所述停车场泊位占用率短时域预测系统包括：
时序数据构建模块，用于基于停车场的历史泊位占用率数据，构建泊位占用率时序数据；
时序数据矩阵构建模块，用于通过时序数据矩阵构建方法，将时序数据拆分为长度相同的时序数据子集，构建时序数据矩阵；
模型训练模块，用于随机选取时序数据矩阵中的数据，结合梯度上升决策树模型，对模型进行训练；
停车场占用率预测模块，用于使用训练后的梯度上升决策树模型对停车场的占用率进行预测。
本发明的另一目的在于提供一种计算机设备，所述计算机设备包括存储器和处理器，所述存储器存储有计算机程序，所述计算机程序被所述处理器执行时，使得所述处理器执行所述的停车场泊位占用率短时域预测方法的步骤。
本发明的另一目的在于提供一种计算机可读存储介质，存储有计算机程序，所述计算机程序被处理器执行时，使得所述处理器执行所述的停车场泊位占用率短时域预测方法的步骤。
本发明的另一目的在于提供一种信息数据处理终端，所述信息数据处理终端用于实现所述的停车场泊位占用率短时域预测系统。
结合上述的技术方案和解决的技术问题，本发明所要保护的技术方案所具备的优点及积极效果为：
本发明提供的停车场泊位占用率短时域预测方法，主要利用停车场后台记录的历史泊位占用率数据，其核心在于使用原始数据，构建时序数据矩阵。首先，获取停车场的历史泊位占用率数据，构建一条占用率时序数据；然后，使用本发明提出的时序数据矩阵构建方法，将原始数据数据构建为一个时序数据矩阵，矩阵中的每一条数据都是原始时序数据的一个子集；其次，随机选取时序数据矩阵中的数据，结合梯度上升决策树模型，对模型进行训练；最后使用训练后的梯度上升决策树模型对停车场的泊位占用率进行预测。本发明将常规泊位占用率预测中使用的一条时序数据，扩充为一个时序数据矩阵，在时序数据矩阵的基础上，随机选取时序数据供泊位占用率预测模型进行训练，从而使模型能更好地学习到泊位占用率变化的规律，提高泊位占用率的预测准确度。
本发明提供了一种基于时序数据矩阵的停车场泊位占用率短时域预测方法，利用停车场收费管理系统后台记录的车辆出入记录，通过数据整理与统计，得到历史泊位占用率数据，使用本发明提出的时序矩阵构建方法，构建出时序数据矩阵作为后续预测方法的将使用到的数据，矩阵中的每一行数据都是一条可供后续预测模型训练与测试的一条相邻时间段内的数据，保证后续随机选取数据时模型能学习到泊位占用率变化的规律，并随机选取矩阵中的数据结合梯度上升决策树模型对模型进行训练和测试，随机选取数据使得模型能学习到整个泊位占用率记录时间段内的泊位占用率变化规律，提高停车场泊位占用率短时域预测的准确度
本发明的技术方案转化后的预期收益和商业价值为：目前各大城市开始推进建设城市级的智慧停车系统，该系统集查询、预约、导航和收费于一体，将该方法嵌入到智慧停车系统为智慧停车预约系统提供更加准确的泊位占用率预测。有益效果为，第一：更准确的泊位占用率预测，用户在进行查询预约车位时，更好地了解其到达目的地附近时停车场的拥挤情况，可以帮助用户根据自身情况更好地选择预约停车位，有效避免出现因预测效果与实际差距较大而影响用户满意度降低的情况，同时帮助用户在停车时避免拥堵和减少寻位时间；第二，更准确的泊位占用率预测，可以帮助系统更好地为用户进行停车场推荐和停车场定价决策，可优先推荐空闲车场，有效地调整区域内停车场泊位占用率的平衡，减少因停车场拥挤而带来的交通拥堵和环境污染。
附图说明
为了更清楚地说明本发明实施例的技术方案，下面将对本发明实施例中所需要使用的附图做简单的介绍，显而易见地，下面所描述的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下还可以根据这些附图获得其他的附图。
图1是本发明实施例提供的停车场泊位占用率短时域预测方法流程图；
图2是本发明实施例提供的时序数据矩阵构建方法将一条长度为T的时序数据构建为列数为N的时序数据矩阵的示意图；
图3是本发明实施例提供的某停车场下使用一条时序数据顺序选取时序数据(常规方法)和时序数据矩阵随机选取时序数据(本发明提出的方法)作为模型的训练集训练后在测试集上的泊位占用率预测结果对比图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
针对现有技术存在的问题，本发明提供了一种停车场泊位占用率短时域预测方法、系统、设备及终端，下面结合附图对本发明作详细的描述。
为了使本领域技术人员充分了解本发明如何具体实现，该部分是对权利要求技术方案进行展开说明的解释说明实施例。
实施例1
如图1所示，本发明实施例提供的停车场泊位占用率短时域预测方法包括以下步骤：
S101，基于停车场的历史泊位占用率数据，构建泊位占用率时序数据；
S102，通过时序数据矩阵构建方法，将时序数据拆分为长度相同的时序数据子集，构建时序数据矩阵；
S103，随机选取时序数据矩阵中的数据，结合梯度上升决策树模型，对模型进行训练；
S104，使用训练后的梯度上升决策树模型对停车场的占用率进行预测。
本发明实施例提供的泊位占用率时序数据，是指按时间顺序构建的等时间间隔的停车场泊位占用率数据，泊位占用率是指某时刻停车场已使用的车位数与总车位数的比值，时间间隔可取任意时长；优选地，时间间隔选取5～180分 钟。
本发明实施例提供的时序数据矩阵构建方法，用于根据一条泊位占用率时序数据，将其拆分为大小相同的时序数据子集，构建一个时序数据矩阵。
本发明实施例提供的时序数据矩阵构建方法，具体为：
假设某停车场的在某天运营时段内泊位占用率时序数据为O＝(O 1,O 2,…,O t,…,O T)，其长度为T，其中O t表示t时刻停车场的泊位占用率值。所述时序数据子集是指长度小于T的泊位占用率时序数据，且时序数据子集中的数据为相邻时刻的泊位占用率，如(O 1,O 2,O 3)为前述时序数据O的子集。建立列数为N的时序数据矩阵，对应的时序数据子集为
Figure PCTCN2022129957-appb-000003
Figure PCTCN2022129957-appb-000004
一共有M＝T-N+1个子集，并将其按行拼接为时序数据矩阵。假设时序数据矩阵的列数为N，其实际意义为使用前N-1个时刻的泊位占用率数据，作为模型输入，预测第N个时刻的泊位占用率，作为模型输出。
本发明实施例提供的梯度上升决策树模型，是指串行生成一系列CART回归树，每一颗树不断拟合上一棵树学习后的残差，逐渐学习到真实值的模型，其训练数据为所述时序数据矩阵中随机选取的数据。
本发明实施例中，在梯度上升决策树模型的训练与测试中，将时序数据矩阵中的数据按任意比例，如7：3，分为训练集和测试集。
本发明实施例提供的停车场泊位占用率短时域预测系统包括：
时序数据构建模块，用于基于停车场的历史泊位占用率数据，构建泊位占用率时序数据；
时序数据矩阵构建模块，用于通过时序数据矩阵构建方法，将时序数据拆分为长度相同的时序数据子集，构建时序数据矩阵；
模型训练模块，用于随机选取时序数据矩阵中的数据，结合梯度上升决策树模型，对模型进行训练；
停车场占用率预测模块，用于使用训练后的梯度上升决策树模型对停车场 的占用率进行预测。
实施例2
作为优选实施例，本发明实施例提供的基于时序数据矩阵的停车场泊位占用率短时域预测方法，利用停车场后台记录的历史泊位占用率数据，构建出时序数据矩阵，并随机选取矩阵中的数据结合梯度上升决策树模型对模型进行训练和预测。其中，本发明实施例提供的时序数据矩阵构建方法为：
如图2所示，假设某停车场的泊位占用率时序数据为O＝(O 1,O 2,O 3,O 4,O 5)，其长度为5，其中O 1＝55％，O 2＝50％，O 3＝65％，O 4＝58％，O 5＝52％。假设时序数据矩阵的列数为3，对应的时序数据子集为
Figure PCTCN2022129957-appb-000005
一共3个子集，并将其按行拼接为矩阵。
为了证明本发明的技术方案的创造性和技术价值，该部分是对权利要求技术方案进行具体产品上或相关技术上的应用实施例。
本发明实施例提供在某省会城市某医院停车在2019.8.1-2020.8.21期间的782362条真实数据值下，分别使用一条时序数据顺序选取时序数据(常规方法)和时序数据矩阵随机选取时序数据(本发明实施例提出的方法)作为模型的训练集训练后，在测试集上的泊位占用率预测结果；本发明实施例分别采用了三种常用的泊位占用率预测模型：线性回归、梯度上升决策树、神经网络，评价指标为平均绝对误差，表达式如下：
Figure PCTCN2022129957-appb-000006
其中M为测试集的中时序数据子集数，
Figure PCTCN2022129957-appb-000007
为模型在t时刻的预测值，y t为车场在t时刻的真实值。
如图3所示，为本发明实施例分别采用三种常用的泊位占用率预测模型的泊位占用率预测结果对比图，可以看到，使用本发明的时序数据矩阵选取训练集，能够减小预测的平均绝对误差，即能够有效提高预测准确度。
本发明实施例在研发或者使用过程中取得了一些积极效果，和现有技术相比的确具备很大的优势，下面内容结合试验过程的数据、图表等进行描述。
下面我们将对本发明提出的方法在测试集带来的效果提升提供简单的数学理论分析：
假设按传统的数据处理方法即按时间先后顺序拆开训练集和测试集，训练集的数据(数据记录时间段内前部分时间的数据)分布服从随机变量
Figure PCTCN2022129957-appb-000008
其中μ 1为X 1的均值，
Figure PCTCN2022129957-appb-000009
为X 1的方差，测试集的数据(数据记录时间段内后部分时间的数据)分布服从随机变量
Figure PCTCN2022129957-appb-000010
传统的数据处理方法，模型学习到的数据分布只是X 1，本发明提出的时序矩阵构建方法，将一条时序数据构建为时序数据矩阵，并随机选取数据为模型提供训练后，模型学习到的数据分布是X 1和X 2的加权平均。
情况1：随机变量X 1和X 2服从相同或相似的分布，即数据记录时间段内前后段时间内的数据变化规律相同或相似，假设那么通过传统数据处理方法和本发明提出的时序数据矩阵，模型训练结果均与X 1具有相同或相似的分布，两者在测试集上的预测结果也相同或相似，其实我们提出的方法效果不差于传统方法；
情况2：随机变量X 1和X 2服从的分布差距很大，可认为X 1和X 2相互独立，假设训练集和测试集的比例为a:b，在传统方法下模型学习的数据分布服从随机变量
Figure PCTCN2022129957-appb-000011
其与测试集的平均绝对误差为：
Figure PCTCN2022129957-appb-000012
在本发明提出的方法下模型学习的数据分布服从随机变量
Figure PCTCN2022129957-appb-000013
Figure PCTCN2022129957-appb-000014
其与测试集(其数据分布也服从
Figure PCTCN2022129957-appb-000015
)的平均绝对误差为：
Figure PCTCN2022129957-appb-000016
即在训练集和测试集中的数据都仅服从一个分布时，理想情况下，我们提出的方法会是预测误差降为0，优于传统方法。
但预测误差一般不为0，因为训练集和测试集中，相同分布的随机变量的比例不一定是相同的，而且这些数据往往都不只服从一个分布，即模型学习到的数据分布只是众多的随机变量的加权平均，在相同分布随机变量集合中测试， 也会存在误差，参考线性回归。而我们提出的方法能有效降低当训练集和测试集的数据服从的随机变量分布不同时带来的误差，使得模型的预测效果更好。
应当注意，本发明的实施方式可以通过硬件、软件或者软件和硬件的结合来实现。硬件部分可以利用专用逻辑来实现；软件部分可以存储在存储器中，由适当的指令执行系统，例如微处理器或者专用设计硬件来执行。本领域的普通技术人员可以理解上述的设备和方法可以使用计算机可执行指令和/或包含在处理器控制代码中来实现，例如在诸如磁盘、CD或DVD-ROM的载体介质、诸如只读存储器(固件)的可编程的存储器或者诸如光学或电子信号载体的数据载体上提供了这样的代码。本发明的设备及其模块可以由诸如超大规模集成电路或门阵列、诸如逻辑芯片、晶体管等的半导体、或者诸如现场可编程门阵列、可编程逻辑设备等的可编程硬件设备的硬件电路实现，也可以用由各种类型的处理器执行的软件实现，也可以由上述硬件电路和软件的结合例如固件来实现。
以上所述，仅为本发明的具体实施方式，但本发明的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本发明揭露的技术范围内，凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等，都应涵盖在本发明的保护范围之内。

Claims (10)

  1. 一种停车场泊位占用率短时域预测方法，其特征在于，所述停车场泊位占用率短时域预测方法包括：利用停车场收费管理系统后台记录的历史泊位占用率数据，构建出时序数据矩阵，并随机选取矩阵中的数据结合梯度上升决策树模型对模型进行训练和预测，确定停车场泊位占用率短时域。
  2. 如权利要求1所述的停车场泊位占用率短时域预测方法，其特征在于，所述停车场泊位占用率短时域预测方法包括以下步骤：
    步骤一，基于停车场的历史泊位占用率数据，构建泊位占用率时序数据；
    步骤二，通过时序数据矩阵构建方法，将时序数据拆分为长度相同的时序数据子集，构建时序数据矩阵；
    步骤三，随机选取时序数据矩阵中的数据，结合梯度上升决策树模型，对模型进行训练；
    步骤四，使用训练后的梯度上升决策树模型对停车场的占用率进行预测。
  3. 如权利要求2所述的停车场泊位占用率短时域预测方法，其特征在于，所述步骤一中的泊位占用率时序数据，是指按时间顺序构建的等时间间隔的停车场泊位占用率数据；泊位占用率是指某时刻停车场已使用的车位数与总车位数的比值，时间间隔取任意时长。
  4. 如权利要求3所述的停车场泊位占用率短时域预测方法，其特征在于，构建泊位占用率时序数据时，所述时间间隔优先选取5～180分钟。
  5. 如权利要求2所述的停车场泊位占用率短时域预测方法，其特征在于，所述步骤二中的时序数据矩阵构建方法，用于根据一条泊位占用率时序数据，拆分为大小相同的时序数据子集，构建一个时序数据矩阵；
    所述时序数据矩阵构建方法具体包括：
    当停车场在运营时段内泊位占用率时序数据为O＝(O 1,O 2,…,O t,…,O T)，长度为T，其中O t表示t时刻停车场的泊位占用率值；所述时序数据子集是指长度小于T的泊位占用率时序数据，且时序数据子集中的数据为相邻时刻的泊位占用率；建立列数为N的时序数据矩阵，对应的时序数据子集为
    Figure PCTCN2022129957-appb-100001
    Figure PCTCN2022129957-appb-100002
    共有M＝T-N+1个子集，并按行拼接为时序数据矩阵。
  6. 如权利要求2所述的停车场泊位占用率短时域预测方法，其特征在于，所述步骤三中的梯度上升决策树模型，是指串行生成一系列CART回归树，每一颗树不断拟合上一棵树学习后的残差，逐渐学习到真实值的模型，其训练数据为所述时序数据矩阵中随机选取的数据；
    在梯度上升决策树模型的训练与测试中，将时序数据矩阵中的数据按任意比例分为训练集和测试集；其中，所述训练集和测试集的比例优选为7：3。
  7. 一种应用如权利要求1～6任意一项所述的停车场泊位占用率短时域预测方法的停车场泊位占用率短时域预测系统，其特征在于，所述停车场泊位占用率短时域预测系统包括：
    时序数据构建模块，用于基于停车场的历史泊位占用率数据，构建泊位占用率时序数据；
    时序数据矩阵构建模块，用于通过时序数据矩阵构建方法，将时序数据拆分为长度相同的时序数据子集，构建时序数据矩阵；
    模型训练模块，用于随机选取时序数据矩阵中的数据，结合梯度上升决策树模型，对模型进行训练；
    停车场占用率预测模块，用于使用训练后的梯度上升决策树模型对停车场的占用率进行预测。
  8. 一种计算机设备，其特征在于，所述计算机设备包括存储器和处理器，所述存储器存储有计算机程序，所述计算机程序被所述处理器执行时，使得所述处理器执行如权利要求1～6任意一项所述的停车场泊位占用率短时域预测方法的步骤。
  9. 一种计算机可读存储介质，存储有计算机程序，所述计算机程序被处理器执行时，使得所述处理器执行如权利要求1～6任意一项所述的停车场泊位占用率短时域预测方法的步骤。
  10. 一种信息数据处理终端，其特征在于，所述信息数据处理终端用于实现如权利要求7所述的停车场泊位占用率短时域预测系统。
PCT/CN2022/129957 2022-09-28 2022-11-04 停车场泊位占用率短时域预测方法、系统、设备及终端 WO2024065954A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211195092.4 2022-09-28
CN202211195092.4A CN115564078A (zh) 2022-09-28 2022-09-28 停车场泊位占用率短时域预测方法、系统、设备及终端

Publications (1)

Publication Number Publication Date
WO2024065954A1 true WO2024065954A1 (zh) 2024-04-04

Family

ID=84742793

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/129957 WO2024065954A1 (zh) 2022-09-28 2022-11-04 停车场泊位占用率短时域预测方法、系统、设备及终端

Country Status (2)

Country Link
CN (1) CN115564078A (zh)
WO (1) WO2024065954A1 (zh)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102024343A (zh) * 2010-12-20 2011-04-20 东南大学 停车场有效泊位占有率短时预测方法
CN109376331A (zh) * 2018-08-22 2019-02-22 东南大学 一种基于梯度提升回归树的城市公交车排放率估计方法
US20200293952A1 (en) * 2019-03-15 2020-09-17 Microsoft Technology Licensing, Llc Categorical feature enhancement mechanism for gradient boosting decision tree
CN112085254A (zh) * 2020-08-05 2020-12-15 中国地质大学（武汉） 基于多重分形协同度量门控循环单元的预测方法及模型
CN113821547A (zh) * 2021-08-26 2021-12-21 中山大学 快速高效的停车场占有率短时预测方法、系统及存储介质
CN114048886A (zh) * 2021-09-15 2022-02-15 深圳大学 基于神经网络的有效泊位组合预测方法、终端及存储介质
CN114579640A (zh) * 2022-03-10 2022-06-03 上海交通大学宁波人工智能研究院 一种基于生成式对抗网络的金融时间序列预测系统和方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102024343A (zh) * 2010-12-20 2011-04-20 东南大学 停车场有效泊位占有率短时预测方法
CN109376331A (zh) * 2018-08-22 2019-02-22 东南大学 一种基于梯度提升回归树的城市公交车排放率估计方法
US20200293952A1 (en) * 2019-03-15 2020-09-17 Microsoft Technology Licensing, Llc Categorical feature enhancement mechanism for gradient boosting decision tree
CN112085254A (zh) * 2020-08-05 2020-12-15 中国地质大学（武汉） 基于多重分形协同度量门控循环单元的预测方法及模型
CN113821547A (zh) * 2021-08-26 2021-12-21 中山大学 快速高效的停车场占有率短时预测方法、系统及存储介质
CN114048886A (zh) * 2021-09-15 2022-02-15 深圳大学 基于神经网络的有效泊位组合预测方法、终端及存储介质
CN114579640A (zh) * 2022-03-10 2022-06-03 上海交通大学宁波人工智能研究院 一种基于生成式对抗网络的金融时间序列预测系统和方法

Also Published As

Publication number Publication date
CN115564078A (zh) 2023-01-03

Similar Documents

Publication Publication Date Title
Lin et al. An efficient deep reinforcement learning model for urban traffic control
Allah Bukhsh et al. Multi-year maintenance planning framework using multi-attribute utility theory and genetic algorithms
Zhang et al. A Traffic Prediction Method of Bicycle-sharing based on Long and Short term Memory Network.
CN108648457A (zh) 一种速度预测的方法、装置和计算机可读存储介质
CN112330215A (zh) 一种城市用车需求量预测方法、设备及存储介质
Muresan et al. Adaptive traffic signal control with deep reinforcement learning an exploratory investigation
CN113287124A (zh) 用于搭乘订单派遣的系统和方法
CN111191839A (zh) 一种换电预测方法和系统及存储介质
CN106850289A (zh) 结合高斯过程与强化学习的服务组合方法
CN113838303B (zh) 停车场推荐方法、装置、电子设备及存储介质
Zamee et al. Self-supervised online learning algorithm for electric vehicle charging station demand and event prediction
WO2024065954A1 (zh) 停车场泊位占用率短时域预测方法、系统、设备及终端
CN116662815B (zh) 时间预测模型的训练方法以及相关设备
Zhou et al. The survival of bike-sharing startups in China: an empirical analysis of the influencing factors
CN116957136A (zh) 一种基于时序深度学习的路面性能预测方法及装置
He et al. The shunting scheduling of EMU first-level maintenance in a stub-end depot
CN115965163A (zh) 基于时空图生成对抗损失的轨道交通短时客流预测方法
Shi et al. Optimal transportation network company vehicle dispatching via deep deterministic policy gradient
CN112434817B (zh) 构建通信算法数据库的方法、装置和计算机存储介质
Lu et al. MOHP-EC: A Multiobjective Hierarchical Prediction Framework for Urban Rail Transit Passenger Flow
Pei Construction and application of talent evaluation model based on nonlinear hierarchical optimization neural network
CN107563555A (zh) 基于Q学习memetic算法的动态多目标软件项目调度方法
Hammedi et al. Two-level optimization to reduce waiting time at locks in inland waterway transportation
Yang An Integrated Framework Integrating Monte Carlo Tree Search and Supervised Learning for Train Timetabling Problem
Li et al. Ridesplitting demand prediction via spatiotemporal multi-graph convolutional network

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22960553

Country of ref document: EP

Kind code of ref document: A1