A dynamic shared parking lot berth resource prediction method

Abstract

The application relates to a dynamic shared parking lot resource prediction method, characterized in that the method comprises the following steps: step S1, calculating a parking lot historical parking lot occupancy rate according to parking lot historical parking data; step S2, constructing a parking lot resource prediction model based on a gated recurrent unit, and predicting a parking lot occupancy rate at a next moment according to parking lot occupancy rate information at a current moment; and step S3, constructing a dynamic shared parking strategy according to a shared parking strategy setting principle and related parameters, combining the parking lot occupancy rate prediction result and the parking lot historical parking data, and outputting a shared time period. Compared with the prior art, the application has the advantages of high prediction accuracy and high universality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of parking lot resource prediction, in particular to a dynamic shared parking lot resource prediction method. BACKGROUND

[0002] With the acceleration of urbanization, the intensification of urban land, and the continuous increase of urban population and car ownership, the problem of urban parking difficulty is becoming increasingly prominent. In recent years, the parking lot sharing policy has emerged in the urban parking management and become an innovative practice to solve the problem of parking difficulty. The main idea of shared parking is to utilize the imbalance of parking demand in time and space to improve the utilization rate of parking lots, thereby effectively alleviating the problem of urban parking difficulty. Setting a reasonable parking lot sharing strategy depends on the prediction of parking lot occupancy rate at future time. Only under the premise of establishing a high-precision parking lot resource demand prediction model, can the dynamic matching of shared parking resources be better achieved.

[0003] Currently, the technology for predicting parking lot occupancy rate at future time mainly has the following shortcomings:

[0004] (1) The prediction method is affected by subjective factors: According to historical documents, the analytic hierarchy process (AHP) is often used to predict parking lot occupancy rate at future time, which is a quantitative analysis method used to solve multi-criteria decision-making problems. Its main process includes determining decision indicators, establishing a hierarchical structure, defining indicator weights, evaluating and analyzing, and predicting, etc. When using the analytic hierarchy process to predict parking lot occupancy rate at future time, the definition of weights needs to be accurate because it will directly affect the accuracy of the prediction results. However, the assignment of weights and priorities in the analytic hierarchy process is based on expert knowledge or user experience, which may be subjective and have a significant impact on data quality, thus there may be certain errors and uncertainties, affecting the accuracy of the prediction results.

[0005] (2) The prediction method is only for single land use: Under different land use, the parking characteristics of each parking lot are different. For example, the parking characteristics of residential areas are that the parking lot occupancy rate is high at night and low during the day; while the parking characteristics of office areas are that the parking lot occupancy rate is low at night and high during the day; in addition, there are differences in the parking characteristics of parking lots in different types of areas such as commercial areas, tourist areas, and hospitals. Therefore, for parking lots in different land use areas, the prediction of parking lot occupancy rate at future time needs to be classified according to the parking demand of the land use, and a statistical model and functional relationship need to be established for each type of parking lot, but this method is too cumbersome and not universal.

[0006] (3) There is still a large prediction error in part of the neural network prediction algorithm: At present, more researchers use machine learning, deep learning and other methods to predict parking lot resources. However, there is still a large error in the prediction results. The prediction algorithm used is old, such as the autoregressive integrated moving average model (ARIMA) proposed in the 1970s, the back propagation (BP) neural network proposed in 1986, the recurrent neural network (RNN) proposed in 1990, the support vector machine model (SVM) proposed in 1995, and the long short-term memory artificial neural network (LSTM) proposed in 1997. These methods can predict parking lot resources at future time, but there is still room for improvement in prediction effect. SUMMARY

[0007] The purpose of the present application is to overcome the defects of the prior art and provide a dynamic shared parking lot resource prediction method with high prediction accuracy and high universality.

[0008] The purpose of the present application can be achieved by the following technical solutions:

[0009] The present application provides a dynamic shared parking lot resource prediction method, which comprises:

[0010] Step S1, calculating the historical parking lot occupancy rate of the parking lot according to the historical parking data of the parking lot;

[0011] Step S2, constructing a parking lot resource prediction model based on a gated recurrent unit, and predicting the parking lot occupancy rate at the next time according to the parking lot occupancy rate information at the current time;

[0012] Step S3, according to the establishment principle and related parameters of the shared parking strategy, combining the parking lot occupancy rate prediction result and the historical parking data of the parking lot, constructing a dynamic shared parking strategy, and outputting the shared time period.

[0013] Preferably, the step S1 includes data cleaning of missing values and abnormal values of the historical parking data of the parking lot.

[0014] Preferably, the input and output of the parking lot resource prediction model based on the gated recurrent unit are:

[0015] Input: parking lot occupancy rate information x at time t tand the hidden state h at time t-1 t-1 ;

[0016] Output: the output y of the hidden node at time t t , representing the parking lot occupancy rate information of the current time; and the hidden state h t .

[0017] Preferably, the hidden state h t The expression is:

[0018]

[0019] wherein, when z t is the update gate, h t-1 is the hidden state at time t-1; is the candidate hidden state, and the expression is:

[0020]

[0021] wherein, h t-1 represents the information of the parking lot occupancy rate contained in the previous time, r t is the reset gate, W and U are weight parameters, b h is a bias parameter, is the Hadamard product symbol, which means element-wise multiplication of matrices, and tanh is the activation function.

[0022] Preferably, a Dropout mechanism is provided in the parking lot resource prediction model based on the gated recurrent unit.

[0023] Preferably, the establishment principle of the shared parking strategy includes maximizing the utilization of parking lot shared resources, the time length of the parking lot open sharing meeting the real needs of users, and the number of parking lot open sharing meeting the real needs;

[0024] The related parameters include the sharable coefficient P share , the maximum sharing coefficient P max , the shortest sharing time T min , the number of shared parking spaces , and the sharing period T share .

[0025] Preferably, the average parking lot occupancy rate calculated according to the historical parking data of the parking lot is represented as the sharable coefficient P share :

[0026]

[0027] In the formula, T is the total number of time points divided by K-minute granularity of historical data; P tis the parking lot occupancy rate of the tth moment.

[0028] Preferably, the shared parking space number T The expression is:

[0029]

[0030] In the formula, A is the total number of parking spaces in the parking lot; is the predicted parking lot occupancy rate of the tth moment; n is the predicted number of vehicles in the parking lot at the tth moment, is the predicted number of vehicles in the parking lot at the tth moment, n

[0031] Preferably, the average parking duration of each parking user in the parking lot historical data is represented as the shortest shared duration T min :

[0032]

[0033] In the formula, l is the amount of parking data of all vehicles in the parking lot historical data; is the entry time of the lth parking data; is the exit time of the lth parking data.

[0034] Preferably, the shared time period T share Specifically:

[0035] According to adjacent moments, m r alternative time windows are formed If the following condition is met , it means that the duration of the alternative time window meets the shortest shared duration, that is, the time window can be shared, and then all alternative time windows that can be shared are selected to form a shared time period.

[0036] Compared with the prior art, the present application has the following beneficial effects:

[0037] 1) The present application combines parking lot historical parking data, uses the gated recurrent unit in the recurrent neural network to predict the parking lot occupancy rate at the future moment, considers factors such as time scale and space scale, determines a dynamic shared parking scheme according to the predicted data, so as to accurately evaluate the parking demand and parking supply, improve the utilization rate of parking resources, improve the efficiency and quality of shared parking services, and alleviate the problem of parking difficulty.

[0038] 2) The parking lot historical parking data is subjected to data cleaning of missing values and abnormal values, which improves the accuracy and reliability of subsequent prediction.

[0039] ​3) Set the Dropout mechanism in the parking space resource prediction model based on the gated recurrent unit, prevent overfitting in the parking lot occupancy rate prediction process by randomly discarding part of the neurons in the Dropout layer;

[0040] 4) The parking space resource prediction model based on the gated recurrent unit has universality, can quickly adapt to new data, has strong flexibility, learns the occupancy rate change characteristics from the historical parking lot occupancy rate of the parking lot, does not need to consider the land use property difference of the parking lot, and can be flexibly used in different land use property areas;

[0041] 5) The sharable coefficient, the maximum sharing coefficient, the shortest sharing time length, the shared parking space number and the sharing time period are used as related parameters of the sharing parking strategy, and accurate output of the sharing time period at the future moment is realized. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The method flowchart of the application;

[0043] Figure 2 The method architecture schematic diagram of the application;

[0044] Figure 3 The input and output structure schematic diagram of the gated recurrent unit;

[0045] Figure 4 The internal structure schematic diagram of the gated recurrent unit;

[0046] Figure 5 The model building schematic diagram of the gated recurrent unit;

[0047] Figure 6 The prediction result diagram of the gated recurrent unit;

[0048] Figure 7 The shared time window selection process schematic diagram;

[0049] Figure 8 The method flowchart in the embodiment. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the application.

[0051] EMBODIMENT

[0052] As Figure 1 and Figure 2As shown, the embodiment gives a dynamic shared parking lot berth resource prediction method, which comprises the following steps:

[0053] Step S1, according to the parking lot historical parking data, the parking lot historical berth occupancy rate is calculated;

[0054] Step S2, a berth resource prediction model based on a gated recurrent unit is constructed, and the parking lot berth occupancy rate at the next moment is predicted according to the parking lot berth occupancy rate information at the current moment;

[0055] Step S3, according to the establishment principle and related parameters of the shared parking strategy, combining the parking lot berth occupancy rate prediction result and the parking lot historical parking data, a dynamic shared parking strategy is constructed, and the shared time period is output.

[0056] Next, the method of the application will be described in detail.

[0057] Step S1 also includes data cleaning of missing values and abnormal values of parking lot historical parking data, which is used to improve the accuracy and reliability of subsequent data analysis and modeling. In this embodiment, the parking lot historical data is analyzed in terms of default value, abnormal value and missing value, so as to ensure the quality and integrity of the data, and the visualization analysis of the parking lot historical data is carried out, so as to intuitively understand the parking rules and characteristics of the parking lot.

[0058] Step S2, a berth resource prediction model based on a gated recurrent unit is constructed, and the parking lot berth occupancy rate at the next moment is predicted according to the parking lot berth occupancy rate information at the current moment, specifically:

[0059] The input and output structure of the gated recurrent unit (such as the attached Figure 3 ):

[0060] Input:

[0061] (1) the input x of t moment t , that is, the parking lot berth occupancy rate information input at the current moment;

[0062] (2) the hidden state h of t-1 moment t-1 , which contains the relevant information of the previous node.

[0063] Output:

[0064] (1) the output y of the hidden node of t moment t , that is, the parking lot berth occupancy rate information output at the current moment;

[0065] (2) the hidden state h passed to the next node t , which contains the relevant information passed to the next node.

[0066] The internal structure of the gated recurrent unit (e.g., FIG. 1) Figure 4 ) :

[0067] The model of the GRU has only two "gates", which can be simply understood as "switches" that determine how much information from the previous time step is retained or ignored. In the context of parking data, Figure 4 where x t represents the parking lot occupancy rate information at time t, h t-1 is the output hidden state information at time t-1, where z t represents the update gate, and r t represents the reset gate. The update gate controls the amount of information from the previous time step that flows into the current time step's parking lot occupancy rate information. The amount of information passed to the next time step is determined by the size of the update gate. If the update gate value is larger, it means that more parking lot occupancy rate information is accepted at this time. The formula for the update gate is as follows:

[0068] z t = σ(W z x t + U z h t-1 + b z ) (1)

[0069] where W z and U z are weight parameters that control the amount of information flow; b z is a bias parameter; and σ represents a sigmoid function that converts the input value to the interval (0, 1).

[0070] The reset gate controls the amount of parking lot occupancy rate information from the previous time step that is ignored. If the reset gate is smaller, it means that more information from the previous time step is ignored. The formula for the reset gate is as follows:

[0071] r t = σ(W r x t + U r h t-1 + b r ) (2)

[0072] where W r and U r are weight parameters that control the amount of information ignored; b r is a bias parameter; and σ represents a sigmoid function.

[0073] The role of the reset gate is to capture short-term dependencies in the sequence, and the role of the update gate is to preserve long-term dependencies in the sequence, in this way, the defect of losing long-term data information in RNN is avoided.

[0074] The formula of the candidate hidden state is:

[0075]

[0076] Where h t-1 represents the information of the parking lot occupancy rate at the previous time, r t represents the reset gate, W and U are weight parameters, b h is the bias parameter, is the Hadamard product symbol, which means element-wise multiplication of matrices, and tanh is the activation function, which can convert data into the interval (-1, 1). When the reset gate r t is closer to 0, it means that more past information is discarded and more current information is left; when the reset gate r t is closer to 1, it means that the past information is more useful. That is, here contains the information of x t input at time t and selectively preserves the hidden layer state h t-1 at time t-1.

[0077] The formula of the final hidden state is:

[0078]

[0079] Where, when z t is closer to 1, it means that the hidden state at the previous time is almost completely forgotten, and the information at the current time is retained to a greater extent; when z t is closer to 0, it means that the hidden state at the previous time is almost completely preserved, and the information at the current time is almost not preserved. The calculation formula of the final hidden state makes the GRU model can effectively handle long-term dependency problems, avoiding the problem of gradient disappearance or gradient explosion in RNN model.

[0080] Model building of gated recurrent unit (as attached Figure 5 ):

[0081] The GRU prediction model mainly consists of an input layer, a hidden layer and an output layer, and also applies the Dropout mechanism to prevent overfitting. First, the data at the current time is input by the input layer, after entering the GRU hidden layer, the hidden layer will process the data features of the current time input and combine the data features of the previous time hidden layer, and then output to the Dropout layer; to prevent overfitting, some neurons will be randomly discarded in the Dropout layer, and then enter the fully connected layer to output the final result.

[0082] GRU prediction instance (as attached Figure 6 ):

[0083] This embodiment selects real data of a community service center parking lot as the prediction case, and the prediction result of the parking lot is shown in the attached Figure 6 It can be found that the predicted value is lower than the real value at almost every time, and the performance is more obvious during the peak period, but this does not have too much impact on the subsequent content, the error is within the acceptable range, and compared with other prediction models, the gap between the predicted value and the real value is relatively small, and the overall upward or downward trend is relatively consistent. At the same time, the prediction error index shows high precision, the RMSE is 0.015650, the MSE is 0.000245, and the MAE is 0.004655. Based on this, it can be concluded that the prediction result obtained by the GRU model has high precision, which has practical significance for determining the shared parking scheme, and can provide effective data support for subsequent decision-making.

[0084] Step three: determine the dynamic shared parking strategy

[0085] According to the prediction result of step three and the historical data of the parking lot, the shared parking strategy is determined

[0086] Principles for establishing the shared parking strategy

[0087] (1) Maximizing the utilization of parking lot shared resources

[0088] The establishment of the shared parking strategy needs to consider how to maximize the utilization of the shared parking resources of the parking lot, while not affecting the original parking demand, so as to improve the utilization efficiency of the parking resources comprehensively.

[0089] (2) The length of time for the parking lot to open for sharing needs to meet the actual needs

[0090] The length of time for the parking lot to share services should be able to meet the actual needs of users. For example, if the length of time for the parking lot to share services is too short, it may not be able to meet the needs of users for long-term parking, thereby reducing the practicality and attractiveness of the shared parking strategy. However, a too long length of time for sharing services may cause waste of parking lot resources and increase the difficulty of management, or may cause too many vehicles to share parking.

[0091] (3) The number of parking spaces for the parking lot to open for sharing needs to meet the actual needs

[0092] When designing the open sharing strategy of the parking lot, the number of parking space supply needs to be considered to ensure the actual needs. The number of parking space supply refers to the number of parking spaces that can be shared in the parking lot. If the shared parking space supply is insufficient, it will lead to the decline of the effect of the shared parking service, and users cannot enjoy sufficient shared resources. Therefore, in order to ensure the feasibility and practicability of the shared parking service, the number of parking spaces of the parking lot open sharing needs to be considered.

[0093] Important parameters of the shared parking strategy:

[0094] (1) Shareable coefficient P share

[0095] In this embodiment, the average parking lot occupancy rate calculated from the historical data of the corresponding parking lot is used as the shareable coefficient.

[0096]

[0097] In the formula: T is the total number of time points divided by 5-minute granularity of historical data; P t is the parking lot occupancy rate at the t time. In this embodiment, it is set to 0.5023.

[0098] (2) Maximum sharing coefficient P max

[0099] In the shared parking strategy, in order to avoid too many shared vehicles from entering the parking lot during the open sharing period, leading to high load of the parking lot and the original parking demand cannot be met, the maximum sharing coefficient index is set in the present invention. This index is set based on historical experience. In this embodiment, 0.85 is used as the maximum sharing coefficient.

[0100] (3) Minimum sharing time T min

[0101] In this embodiment, the average parking time of each parking user in the historical data of the corresponding parking lot is used as the minimum sharing time:

[0102]

[0103] In the formula: l is the amount of parking data of all vehicles in the historical data of the parking lot; is the entry time of the lth parking data; is the exit time of the lth parking data. In this embodiment, it is 3414.88 seconds.

[0104] (4) Shared parking space number

[0105] It is necessary to determine how many shared parking spaces can be provided at a certain time in the future for the implementation of the sharing strategy. Therefore, the present application introduces the important parameter of the number of shared parking spaces, which can help users who come to participate in shared parking to understand the information about the number of remaining shared parking spaces, and also provide help for further screening of shared time periods.

[0106]

[0107] In the formula: A is the total number of parking spaces in the parking lot; is the predicted parking space occupancy rate at the t n time; is the predicted number of vehicles in the parking lot at the t n time,

[0108] The number of shared parking spaces at each time in this embodiment is shown in Table 1:

[0109] Table 1

[0110]

[0111] (5) Shared time period T share

[0112] According to the adjacent times, m r alternative time windows can be formed If , it means that the length of the alternative time window meets the minimum sharing length, that is, the time window can be used for sharing. In this way, all alternative time windows that can be used for sharing are selected to form a shared time period, which is composed of m shared time windows [t open , t close ], as shown in the attached Figure 7 . Outputting the shared time window is the most important part of the present shared parking strategy.

[0113] The shared time period information output by this embodiment is shown in Table 2, and there are four segments that meet the sharing conditions:

[0114] Table 2

[0115] Start time End time Period length Minimum shared duration Whether the sharing condition is met 2021 / 11 / 13 00:00 2021 / 11 / 13 08:00 28800s 3414.88s Yes 2021 / 11 / 13 11:55 2021 / 11 / 13 12:00 300s 3414.88s No 2021 / 11 / 13 12:05 2021 / 11 / 13 12:40 2100s 3414.88s No 2021 / 11 / 13 16:05 2021 / 11 / 14 09:30 62700s 3414.88s Yes 2021 / 11 / 14 10:10 2021 / 11 / 14 13:30 12000s 3414.88s Yes 2021 / 11 / 14 13:40 2021 / 11 / 14 14:30 3000s 3414.88s No 2021 / 11 / 14 14:40 2021 / 11 / 15 00:00 33600s 3414.88s Yes

[0116] To sum up, the application adopts a prediction method based on historical data, which avoids the influence of subjective factors on the prediction results in the traditional method. Secondly, the prediction model has high universality and can be widely applied to different land use scenarios. Finally, the application adopts a gated recurrent unit (GRU) model, which can improve the prediction efficiency and accuracy of the trained model, effectively solving the problem of limited model prediction accuracy in the prior art. Therefore, the method proposed in the application can effectively evaluate the parking demand and parking supply, thereby improving the quality and efficiency of shared parking services.

[0117] The above merely describes specific embodiments of the application, but the protection scope of the application is not limited thereto, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed in the application, which should be covered in the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. A method for predicting resources of parking spaces in a dynamic shared parking lot, characterized in that, The method comprises: Step S1, calculating a parking lot historical berth occupancy rate according to parking lot historical parking data; Step S2, constructing a berth resource prediction model based on a gated recurrent unit, and predicting a next-time parking lot berth occupancy rate according to current-time parking lot berth occupancy rate information; Step S3, constructing a dynamic shared parking strategy according to a shared parking strategy setting principle and related parameters, combining the parking lot berth occupancy rate prediction result and the parking lot historical parking data, and outputting a shared time period; The shared parking strategy setting principle comprises maximization utilization of parking lot berth shared resources, time length of parking lot opening sharing meeting real needs of users, and berth quantity of parking lot opening sharing meeting real needs; The related parameters include a sharable coefficient , a maximum sharable coefficient , a shortest sharing duration , a number of shared parking spaces , and a sharing time period ​ The average parking lot occupancy rate calculated according to the historical parking data of the parking lot is characterized as a shareable coefficient : （1） In the formula, is the total number of historical data divided by K-minute granularity; is the parking lot occupancy rate at the time. The number of shared parking spaces The expression is: （2） In the formula: is the total number of parking spaces of the parking lot; is the predicted parking space occupancy rate of the parking lot at the time; is the predicted number of vehicles in the parking lot at the time, ; The average parking duration of each parking user in the parking lot historical data is represented as the shortest shared duration : （3） In the formula: This represents the total number of vehicles parked in the historical data of this parking lot. For the first The entry time of each parking data point; For the first The time of exit for each parking data point; The sharing period Specifically: According to adjacent Moment, the composition Alternative time window If the condition , which means that the duration of the alternative time window meets the shortest sharing duration, that is, the time window can be used for sharing, then all alternative time windows that can be used for sharing are selected to form a sharing period.

2. The method of claim 1, wherein, The step S1 comprises data cleaning of missing values and abnormal values of the parking lot historical parking data.

3. The method of claim 1, wherein, The input and output in the berth resource prediction model based on the gated recurrent unit are specifically; Output: parking lot occupancy information at the time and hidden state at the time ; Output: the output of the hidden node at the time instant representing the parking lot occupancy information output at the current time instant; and the hidden state passed to the next node .

4. The method of claim 3, wherein, The hidden state The expression is: （4） wherein, when is the update gate, is the hidden state at time t; is the candidate hidden state, expressed as: （5） wherein, represents information of the parking lot occupancy rate contained at the previous time, for resetting the door, and is a weight parameter, is a bias parameter, is a Hadamard product symbol, indicating element-wise multiplication of matrices, is an activation function.

5. The method of claim 1, wherein, The berth resource prediction model based on the gated recurrent unit is provided with a Dropout mechanism.

Images