Planning method, system and device for optimal path of autonomous parking in parking lot

Abstract

The present invention belongs to the field of transportation, and specifically relates to a method for planning the optimal path for autonomous parking in a parking lot and a corresponding parking guidance system and device in the parking lot. The scheme first uses the GCN graph convolutional network and the GRU layer to construct a T-GCN time graph convolutional network. Then, the historical data of the parking lot is encoded into a specific format and used to train the T-GCN time graph convolutional network, thereby obtaining a network model that can realize the prediction of vehicle flow and parking space distribution. Next, the vehicle flow and parking space prediction network is used to predict the information of vacant parking spaces in the parking lot, and then the predicted time to each candidate parking space is calculated in combination with the parking lot parking model. Finally, the candidate routes are sorted in ascending order of time and fed back to the driver for the driver to choose. The scheme provided by the present invention can assist users in completing the decision-making behavior of parking in parking spaces, save users' time and energy in finding parking spaces, and improve users' parking efficiency and user experience in parking lots.

Description

Technical Field

[0001] The present invention belongs to the field of transportation, and in particular relates to a method for planning an optimal path for autonomous parking in a parking lot and a corresponding guidance system and device for parking in the parking lot. Background Art

[0002] Existing large shopping malls and office buildings are equipped with underground parking lots. The access system in underground parking lots is relatively complex. Although this can provide drivers with more options for parking routes, it also makes it difficult for drivers to choose. In addition, with the continuous improvement of social and economic levels, the number of motor vehicles in society is also increasing, which makes drivers' demand for parking lots higher and higher. For this reason, as more and more motor vehicles enter and exit the parking lot, there is more uncertainty about the available parking spaces in the parking lot.

[0003] Most drivers consider time and convenience as their main needs in daily travel. Therefore, how to choose the best parking route in the parking lot has become a difficult problem. Although some parking lots will remind users of the number of remaining parking spaces in the lot when entering and exiting, they cannot accurately inform drivers of the spatial location of the vacant parking spaces in the parking lot. This causes users to spend a lot of time looking for vacant parking spaces in the parking lot after entering the parking lot, which not only wastes users' time and energy, but may also aggravate the congestion of the parking lot. Under this circumstance, helping users predict the spatial distribution of vacant parking spaces in the parking lot in advance and providing guidance for users' parking space selection and route planning have become the main research directions for improving parking lot management efficiency and enhancing users' parking experience in the parking lot.

[0004] The problem of optimal parking path planning and selection in a parking lot is essentially a problem of calculating the parking time for multiple candidate parking spaces. In this case, the optimal parking route should be flexibly adjusted in real time according to the dynamic release of parking spaces in the parking lot. However, the existing various types of driver-side equipment or software services can only display the location of the parking lot, but cannot grasp the number and location information of the vacant parking spaces in the parking lot; this greatly increases the difficulty of optimal parking path planning in the parking lot. How to improve parking efficiency in this special situation is becoming a difficult problem that technicians in this field need to solve urgently. Summary of the invention

[0005] In order to solve the problem that it is difficult to implement optimal parking path planning in a parking lot in the prior art and the traffic efficiency of the planning scheme is difficult to guarantee, the present invention provides a planning method for an optimal parking path for autonomous parking in a parking lot, and a corresponding parking guidance system and device in the parking lot.

[0006] The technical solution provided by the present invention is:

[0007] A method for planning an optimal path for autonomous parking in a parking lot is used to combine the prediction results of the parking space occupancy status in the parking lot and the spatial distribution of the parking spaces to assist the user in the parking decision-making process. The method includes the following steps:

[0008] S1: Obtain the historical status data set of the specified parking lot. The feature data in the historical status data set includes: a map of the distribution of parking spaces in the parking lot; G , representing the number of parking spaces in the parking lot t The parking space state matrix of the occupancy state at the time C t , and characterize parking lots in t Traffic flow feature vector of traffic flow information at the moment S t .

[0009] S2: Use the historical status dataset of each parking lot to train a T-GCN temporal graph convolutional network consisting of GCN layers and GRU layers.

[0010] The trained network model is used as a parking space status prediction network for predicting the occupancy status of each parking space in the parking lot at a future time.

[0011] S3: The vehicle reaches the designated parking space from the current position i Expected time T i To optimize the goal, a parking space selection decision model is established with the idle status of each parking space as a constraint.

[0012] S4: When any user makes a parking lot planning request, the recommended parking spaces are generated and returned to the user through the following steps:

[0013] (1) The historical status dataset of each parking lot in the request area is input into the parking space status prediction network, and a parking space status matrix sequence representing the occupancy status of each parking space at a future moment is output.

[0014] (2) Combined with the dynamically updated parking space status matrix of the parking lot at future times, the decision model for parking space selection is solved to obtain the expected time for the user to reach each vacant candidate parking space. T i .

[0015] (3) According to the expected time T i Sort the candidate parking spaces from short to long, and return the top-ranked parking spaces to the user as recommended parking spaces;

[0016] S5: Based on the current location of the user included in the request information, generate navigation information to the ideal parking space selected by the user, and return it to the user.

[0017] As a further improvement of the present invention, in step S1, the map representing the map information inside the parking lot G The data format is: G =( V , E ),in, V is the set of nodes corresponding to parking spaces in the parking lot, V={ v 1 , v 2 , …, v n}, v i Indicates the number of i parking spaces; E Represents a collection of paths between parking spaces in a parking lot.

[0018] Parking space status matrix C t Any element in c i Indicates the corresponding parking space in the parking lot v i exist t The occupancy status at the moment.

[0019] Traffic flow feature vector S t The data format is: S t =( d t , q t , v t , p t , U t ) T .in, d t Indicates the current day of the week; d t ∈{1,2,…,7}; q t Indicates the current parking lot traffic volume; v t Indicates the vehicle flow speed in the parking lot at the current moment; p t Indicates the location information of the available parking spaces in the parking lot at the current moment; U tIndicates the time required for the vehicle to leave the parking lot at the current moment.

[0020] As a further improvement of the present invention, in step S2, the T-GCN temporal graph convolutional network includes an input layer, a GCN layer, a GRU layer and an output layer.

[0021] The input layer is used to C t and S t A simple check of the sequence data was performed and compared with the G The GCN layer learns complex topological structures through graph convolutional networks, captures spatial features between nodes in the graph G, and stacks multiple convolutional layers to obtain spatial correlation. The GRU layer is used to obtain the temporal features of the C t and S t The time series of the input gated recurrent unit model is used to obtain dynamic changes through information transmission between units, and the temporal characteristics of the traffic flow information are obtained. The output layer is used to predict and output the parking space state matrix at the future moment based on the captured spatial and temporal features.

[0022] As a further improvement of the present invention, the state equation of the GCN layer in the T-GCN temporal graph convolutional network is as follows:

[0023] ,

[0024] In the above formula, Represents the matrix after preprocessing the self-connection matrix; is the self-connection matrix of A, which is used to connect the nodes represented by each parking space and satisfies: ; I N is the identity matrix, is an adjacency matrix, which represents the connection relationship between parking space nodes and satisfies: , the elements in the matrix is 0 or 1, indicating the existence or non-existence of an edge between nodes; yes The output of the layer reflects the feature information between parking spaces; It is the parking space feature information after passing through the GCN layer. is a parameter containing the layer; Represents the sigmoid function; represents the weight matrix from the input layer to the hidden layer, P is the length of the feature matrix, including parking space status features, traffic flow information features, and parking lot map features. H is the number of hidden units; Represents the weight matrix from the hidden layer to the output layer. Represents the predicted length T The output of represents the updated features of each parking space node; LU is the activation function.

[0025] As a further improvement of the present invention, the state equation of the GRU layer in the T-GCN temporal graph convolutional network is as follows:

[0026] ,

[0027] In the above formula, h t is the hidden state of the current node; represents the candidate hidden state; h t-1 The hidden state passed from the previous node; z i represents the output of the update gate; y i is the output of the current hidden node; Represents the sigmoid function; W 0 represents the initial weight matrix; x t is the current output; r t Represents the output of the reset gate; tanh is the hyperbolic tangent activation function; W z , W r , They are all concatenated vector weight matrices, which are also the parameters that the GRU model needs to learn.

[0028] As a further improvement of the present invention, in step S3, the expected time T i Travel time through curves t 1i , Straight-line driving time t 2i and parking time t 3i It consists of three parts: Curve driving time t 1i The calculation formula is as follows:

[0029] ,

[0030] In the above formula, s 1i represents the turning distance; v1 represents the maximum safe speed of the vehicle when driving on a curve in the parking lot; R represents the turning radius of the vehicle; Indicates the number of bends w i The calculated curve rotation angle; when w i =1, then =π / 2, when w i =2, then =π, and so on.

[0031] As a further improvement of the present invention, the straight-line driving time t 2i The calculation formula is as follows:

[0032] ,

[0033] In the above formula, s 2i Indicates the total distance traveled in a straight line; s 0 Indicates the speed change driving distance before and after each curve; v 2 Indicates the uniform driving speed on a straight road section; s 0 Indicates the speed change time before and after each curve.

[0034] As a further improvement of the present invention, the parking time t 3i The calculation formula is as follows:

[0035] ,

[0036] In the above formula, Indicates the parking space type. =1, 2, 3 correspond to lateral parking spaces, diagonal parking spaces, and vertical parking spaces respectively; t 1 , t 2 and t 3 Respectively represent the average parking time for lateral parking spaces, diagonal parking spaces and vertical parking spaces.

[0037] The present invention also includes a parking guidance system in a parking lot, which adopts the aforementioned parking lot autonomous parking optimal path planning method to provide parking space recommendation and navigation services to the user when the user issues a parking request. The parking guidance system in the parking lot provided by the present invention includes: a data acquisition unit, a parking space state prediction network, a parking space dynamic release unit, a parking time calculation unit, a parking space recommendation unit, and a navigation unit.

[0038] The data collection unit is used to obtain the map information of the parking space distribution in each parking lot, and collect the parking space occupancy status and vehicle flow information in and out of the parking lot at each moment in real time. G、 Parking space status matrix C t and traffic flow feature vector S t .

[0039] The parking space status prediction network is trained by the T-GCN temporal graph convolutional network. The parking space status prediction network is used to predict the current parking lot based on the input graph. G And the parking space status matrix C t and traffic flow feature vector S t The historical sequence data is used to output the predicted parking space status matrix at the future moment. C t sequence data.

[0040] The parking space dynamic release unit is used to obtain the future parking space status matrix output by the parking space status prediction network in real time. C t The available parking spaces in the parking lot are dynamically updated.

[0041] The parking time calculation unit is used to calculate the time required for the three stages of driving, namely, curve, straight line and parking, starting from the parking lot entrance and ending at each candidate parking space. t 1i , t 2i , t 3i , add the three together to get the expected time for each parking space T i .

[0042] The parking space recommendation unit is used to calculate the occupancy status of each parking space in the parking lot. c i is the decision coefficient, and each available free parking space is selected according to the expected time T i Sort in ascending order, generate a corresponding candidate list and send it to the user.

[0043] The navigation unit is used to provide navigation services to the user based on the user's current location and the parking space selected from the candidate list, and guide the user to complete parking.

[0044] The present invention also includes a parking guidance device in a parking lot, which includes a memory, a processor, and a computer program stored in the memory and running in the processor. When the processor executes the computer program, the aforementioned optimal path planning method for autonomous parking in the parking lot is adopted to provide parking space recommendation and navigation services to the user when the user issues a parking request.

[0045] The technical solution provided by the present invention has the following beneficial effects:

[0046] The present invention aims at the situation that the occupancy status of parking spaces is related to the traffic flow of the parking lot and the user's preference, and the traffic flow has a regularity in the time domain, and the user's choice of parking spaces has a regularity in the spatial domain; a T-GCN temporal graph convolutional network is constructed using the GCN graph convolutional network and the GRU layer. The T-GCN temporal graph convolutional network is used to learn and capture the temporal and spatial features in the historical management data of the parking lot, and then a network model that can realize the prediction of traffic flow and parking space distribution is obtained.

[0047] The present invention uses a parking space status prediction network trained based on the T-GCN temporal graph convolutional network to predict the status of vacant parking spaces in a parking lot in multiple time periods in the future, thereby helping drivers to implement optimal parking path planning in the parking lot, solving the problems that existing research cannot effectively reflect the distribution of vacant parking spaces in multiple time periods in the future and the low prediction accuracy, and can improve parking efficiency.

[0048] The present invention also combines a new vehicle prediction network to design a general method for realizing optimal parking path planning in a parking lot. The method first uses the traffic flow and parking space prediction network to predict the information of the available parking spaces in the parking lot, and then combines the parking model of the parking lot to calculate the predicted time to each candidate parking space. Finally, the candidate routes are sorted in ascending order of time and fed back to the driver for the driver to choose. The solution of the present invention assists users in completing the decision-making behavior of parking in parking spaces, saves users' time and energy in finding parking spaces, and improves users' parking efficiency and user experience in parking lots. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flowchart of the steps of the method for planning the optimal path for autonomous parking in a parking lot provided in Example 1 of the present invention.

[0050] Figure 2 This is a model architecture diagram of the parking space status prediction network constructed based on the T-GCN temporal graph convolutional network in Example 1 of the present invention.

[0051] Figure 3 This is a network module diagram of the GCN layer used in Example 1 of the present invention.

[0052] Figure 4 This is a network module diagram of the GRU layer used in Example 1 of the present invention.

[0053] Figure 5 This is a spatial layout diagram of a small parking lot that includes lateral parking spaces and vertical parking spaces.

[0054] Figure 6 This is a spatial layout diagram of a small parking lot with diagonal parking spaces.

[0055] Figure 7 This is a module diagram of a parking guidance system in a parking lot provided in Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0057] Example 1

[0058] The technicians in this case found that in the daily management of parking lots, the occupancy status of parking spaces is mainly related to two factors. One is the traffic flow in the parking lot, and the other is the group rules implied in the user's personal preferences for parking space selection. Among them, the traffic flow mainly affects the total number of vacant parking spaces in the parking lot. This is reflected in: when the number of vehicles entering the parking lot is greater than the number of vehicles leaving, the vacant parking spaces in the parking lot will be gradually occupied. Conversely, when the number of vehicles entering the parking lot is less than the number of vehicles leaving, the vacant parking spaces in the parking lot will be gradually released. Considering that the daily changes in the traffic flow in the parking lot are related to the user's travel, the changes in the traffic flow will show obvious rules in the time domain. For example: for a parking lot in an office building, the traffic flow will increase significantly during the morning and evening rush hours on weekdays, while for a parking lot in a shopping mall-type building, the traffic flow will increase significantly during holidays, and so on.

[0059] The user's personal preference for parking space selection will affect the occupied status of each parking space. For example, most users tend to park their vehicles in a location where there are no vehicles parked on both sides; when there is only one free parking space around, most users prefer to park their vehicles in a parking space close to a pillar and close to a non-solid wall. Among the three typical parking spaces, lateral parking spaces, diagonal parking spaces, and vertical parking spaces, most users prefer to park their vehicles in diagonal parking spaces. In short, the spatial attributes of the free parking spaces themselves and the spatial distribution of the already parked parking spaces will affect the probability of the remaining free parking spaces being selected by users. That is, the probability of each parking space being selected and occupied by users will show obvious regularities in the spatial domain.

[0060] Based on the above findings, the technical personnel in this case provided a planning method for the optimal path for autonomous parking in a parking lot. This method can combine the predicted results of the parking space occupancy status in the parking lot and the spatial distribution of the parking spaces to assist the user in the decision-making process of parking behavior when entering the parking lot.

[0061] The method provided in this embodiment includes two stages. In the first stage, a neural network model is trained through the historical management data of the parking lot, so that the network model can learn the time characteristics contained in the historical traffic flow of the parking lot, and the spatial characteristics contained in the decision-making of users to choose parking spaces under different conditions. Then, a network model is obtained that can predict the occupancy of each parking space in the parking lot at a future time based on the input parking space layout of the parking lot and the historical data of parking space occupancy status and traffic flow. In the second stage, the number and spatial distribution of free parking spaces in the parking lot are predicted in real time by completing the training network model, and then the time for users to arrive at each free parking space is calculated based on this constraint. Finally, parking spaces are recommended to users in the priority order of the shortest driving time, and users are allowed to choose independently.

[0062] Specifically, Figure 1 As shown, the method for planning the optimal path for autonomous parking in a parking lot provided in this embodiment includes the following steps:

[0063] S1: Obtain a historical status data set of a designated parking lot, which is used to train the network model. The feature data in the historical status data set of this embodiment includes: a map information representing the distribution of parking spaces in the parking lot; G , representing the number of parking spaces in the parking lot t The parking space state matrix of the occupancy state at the time C t , and characterize parking lots in t Traffic flow feature vector of traffic flow information at the moment S t .

[0064] In the solution of this embodiment, in order to facilitate the learning of the neural network model, the map information representing the distribution of parking spaces in the parking lot is converted into an unweighted graph structure data, denoted as graph G .picture G Each node in the graph is a parking space in the parking lot, and the edges between nodes can be used to represent the internal passages of the parking lot. G The data format is: G =( V , E ),in, V is the set of nodes corresponding to parking spaces in the parking lot, V={ v 1 ,v 2 , …, v n}, v i Indicates the number of i parking spaces; E Represents a collection of paths between parking spaces in a parking lot.

[0065] This embodiment uses a parking space status matrix to represent the idle state of each parking space in the parking lot at a specified time. Specifically, the parking space status matrix C t Any element in c i Indicates the corresponding parking space in the parking lot v i exist t On this basis, the parking space state matrix C t As a figure G By adding the weight information of the parking lot, a weighted graph reflecting the spatial distribution of vacant parking spaces or occupied parking spaces can be obtained. .

[0066] The traffic flow feature vector is a state data reflecting the change of the traffic flow in the parking lot, which contains the values ​​of several characteristic parameters that can characterize the vehicles after optimization. S t The data format is:

[0067] S t =( d t , q t , v t , p t , U t ) T .

[0068] in, d t Indicates the current day of the week; d t ∈{1,2,…,7}; q t Indicates the current parking lot traffic volume; v t Indicates the vehicle flow speed in the parking lot at the current moment; p t Indicates the location information of the vacant parking spaces in the parking lot at the current moment; Ut It should be noted that in other embodiments, in order to optimize the prediction accuracy and generalization of the model, the traffic flow feature vector S t The types and quantities of characteristic parameters in can be flexibly adjusted as needed, such as appropriately reducing parameters or adding new parameters, etc.

[0069] S2: Use the historical status dataset of each parking lot to train a T-GCN temporal graph convolutional network so that the network model can learn the parking space status matrix at several moments in the future. Mapping function between historical data and feature data f .Right now:

[0070] ,

[0071] In the above formula, m represents the dimension of the model input and output data.

[0072] The trained network model is used as a parking space status prediction network for predicting the occupancy status of each parking space in the parking lot at a future time.

[0073] In the solution of this embodiment, the required parking space status prediction network is a typical tool for predicting future status data based on past historical status data. Considering that the parking space status matrix in this embodiment needs to be predicted by combining the temporal features and spatial features in the historical data, the technicians combined GCN (Graph Convolutional Network) and GRU (Gate Recurrent Unit) as the backbone network to construct the following Figure 2 The T-GCN temporal graph convolutional network shown.

[0074] Figure 2 The T-GCN temporal graph convolutional network includes an input layer, a GCN layer, a GRU layer, and an output layer. The input layer is used to obtain input data and C t and S t of m A simple check is performed on the dimensional sequence data and compared with the graph G The parking space node V Passageway between parking spaces E The information is input into the GCN layer together. The data verification process of the input layer includes identifying and removing duplicate data and missing elements.

[0075] The GCN layer learns complex topological structures through graph convolutional networks, captures the spatial features between nodes in the graph G, and uses multiple convolutional layers to stack to obtain spatial correlation. Specifically, the network structure of the GCN layer in this embodiment is as follows: Figure 3 As shown. Figure 3 It can be seen that the GCN layer takes a multi-dimensional feature with multiple input units as input, passes through the middle hidden layer, and obtains the output of multiple output units.

[0076] Assume that an adjacency matrix is ​​given A and the feature matrix X , the model of the GCN layer constructs a filter in the Fourier domain. This filter acts on the nodes of the feature map and captures the spatial features between nodes through its first-order neighborhood. The process can be expressed as:

[0077] ,

[0078] In the above formula, Represents the matrix after preprocessing the self-connection matrix; is the self-connection matrix of A, which is used to connect the nodes represented by each parking space and satisfies: ; I N is the identity matrix, yes The output of the layer reflects the feature information between parking spaces; It is the parking space feature information after passing through the GCN layer. is a parameter containing the layer; Represents the sigmoid function; is an adjacency matrix, which represents the connection relationship between parking space nodes and satisfies:

[0079] ,

[0080] Elements in the matrix It is 0 or 1, indicating the existence or non-existence of an edge between nodes.

[0081] The model of the GCN layer selects two convolutional layers to capture spatial dependencies. The process can be expressed as:

[0082] ,

[0083] In the above formula, represents the weight matrix from the input layer to the hidden layer, P is the length of the feature matrix, including parking space status features, traffic flow information features, and parking lot map features. H is the number of hidden units; Represents the weight matrix from the hidden layer to the output layer. Represents the predicted length T The output of represents the updated features of each parking space node; ReLU is the linear rectification function used in the activation layer.

[0084] Combination Figure 3 ,Through the above changes, the GCN layer can realize the C t , S t Output Z through the hidden layer 1 , Z 2 , Z 3 , Z 4 , represents the parking space node set features, parking space passage set features, parking space state features and vehicle flow features after processing by the GCN layer, and is represented by Y i Add label information to each output feature.

[0085] The GRU layer is used to obtain the temporal features C t and S t The time series of the input gated recurrent unit model is used to obtain dynamic changes through information transmission between units, and the temporal characteristics of the traffic flow information are obtained.

[0086] In this embodiment, if Figure 4 As shown in the figure, the model structure of the GRU layer is relatively simple and has fewer parameters. Unlike the LSTM long short-term memory network, the GRU unit only contains an update gate and a reset gate. Therefore, when this model is selected to obtain time correlation, the model has a faster training capability.

[0087] Among them, the transfer function of the update gate in the GRU layer is as follows:

[0088] ,

[0089] In the above formula, z t Represents the output of the update gate, which is used to control the update and retention of parking lot information, helping the model decide whether to introduce new parking lot data or keep the previous prediction information; x t is the current output, indicating the parking space status information and vehicle flow information at the current time step; h t-1 It is the hidden state passed down from the previous node, including the parking lot state of the previous time step; yes s Type activation function, , used to ensure the update ratio of feature information; Wz is the parameter that the GRU layer needs to learn, which is the concatenated vector weight matrix and satisfies: W z = W zx + W zh .

[0090] in, W zx Is with the current input x t The relevant weight matrix is ​​used to map the parking lot status data and parking lot traffic data to the space related to the update gate. W zh is the hidden state at the previous moment h t-1 The relevant weight matrix is ​​used to map the hidden state (including historical information) of the previous moment to the space related to the update gate calculation. It helps the model to calculate the current update gate by combining the previous state information.

[0091] The transfer function of the reset gate in the GRU layer is as follows:

[0092] ,

[0093] In the above formula, r t Represents the output of the reset gate, which determines how to reset the hidden state of the previous moment h t-1 With the current input x t Combined, the value of the reset gate is between 0 and 1, with 0 indicating a complete reset and 1 indicating a complete retention. h t-1 It is the hidden state passed down from the previous node; W r is the parameter that the GRU layer needs to learn, which is the concatenated vector weight matrix and satisfies: W r = W rx + W rh .

[0094] in, W rx Is with the current input x t The relevant weight matrix maps the parking lot status data and parking lot traffic data to the space of the reset door. W rh is the hidden state at the previous moment ht-1 The relevant weight matrix converts the previous hidden state h t-1 Mapped to the space of the reset gate. The weight matrix of each part is responsible for processing the input and historical information respectively, and synthesizing this information through the activation function of the reset gate to decide how to reset the hidden state.

[0095] Candidate hidden states The transfer function is as follows:

[0096] ,

[0097] in, x t is the current output, h t-1 is the hidden state passed down from the previous node, tanh is the activation function, and the data is scaled to between -1 and 1. r t represents the output of the reset gate; tanh is the hyperbolic tangent activation function, ; is the parameter that the GRU model needs to learn, which is the concatenated vector weight matrix and satisfies: .

[0098] in Indicates that the current input x t The associated weight matrix maps the input to the space of candidate states, Represents the hidden state at the previous moment h t-1 The associated weight matrix maps the state to the space of candidate states. In actual calculation, the two parts are matrix added and then the hyperbolic tangent activation function is applied.

[0099] The output layer is used to predict and output the parking space state matrix at future moments based on the captured spatial features and temporal features. Specifically, the output layer in the T-GCN temporal graph convolutional network of this embodiment outputs a set Z of occupancy states of each parking space predicted in multiple time periods in the future for the specified parking lot:

[0100] Z=[ C t+1 ,…, C t+m ],

[0101] in, C t+1 The parking space state matrix represents the occupancy state of each parking space in the parking lot at time t+1.

[0102] S3: The vehicle reaches the designated parking space from the current position iExpected time T i To optimize the goal, a parking space selection decision model is established with the idle status of each parking space as a constraint.

[0103] ,

[0104] In the above formula, c i Indicates the parking lot i The occupancy status of parking spaces, i=1… n , n Indicates the total number of parking spaces in the parking lot; c i =1 means the parking space is vacant, c i =0 means the parking space is occupied. t 1i , t 2i and t 3i Respectively indicate that the user reaches i The time for driving on curves, driving in a straight line and parking in a parking space are calculated.

[0105] In this embodiment, the parking space recommendation problem is transformed into a single-objective optimization problem of finding a parking space that is idle at a specified time and can meet the goal of minimizing the user's arrival time, and a data target for the problem, namely the aforementioned decision model, is established.

[0106] In the subsequent process, according to the parking space status matrix at future moments predicted by the parking space status prediction network, the parking space release data in the parking lot at future moments can be dynamically updated, and then the above decision model is solved according to each released parking space to determine the candidate parking spaces that meet the optimization goal.

[0107] It should be noted that in other embodiments, new attributes can be added to each parking space according to the special requirements of each user for the parking space, and other constraints can be added to the decision model. To indicate the parking space type, =1, 2, 3 can correspond to lateral parking spaces, diagonal parking spaces, and vertical parking spaces respectively. If a user only wants to park the vehicle in a diagonal parking space, then a constraint condition can be added to the user's decision model: =2. The corresponding decision model is changed to:

[0108] ,

[0109] S4: When any user makes a parking lot planning request, the recommended parking spaces are generated and returned to the user through the following steps:

[0110] (1) Request a map of each parking lot in the area G And the parking space status matrix C t and traffic flow feature vector S t The historical data is input into the parking space status prediction network, and a parking space status matrix sequence representing the occupancy status of each parking space at the future moment is output.

[0111] (2) Combined with the dynamically updated parking space status matrix of the parking lot at future times, the decision model for parking space selection is solved to obtain the expected time for the user to reach each vacant candidate parking space. T i .

[0112] In the solution of this embodiment, it is considered that the distance from any position of the user to different parking spaces in the parking lot can be divided into two sections, one of which is the distance from the current position to the garage entrance, and the other is the distance from the garage entrance to the parking space. Among them, for different selected parking spaces, the former is actually the same, while the latter is different. Therefore, the expected time in the optimization objective of the decision model in this embodiment is T i It actually refers to the expected time for users to get from the garage entrance to each parking space in the parking lot.

[0113] For the convenience of calculation, this embodiment further divides the distance that users need to travel to each parking space into a straight-line distance, a curved distance, and a parking entrance stage according to different parking space conditions. T i It is divided into three parts: i Curve driving time for parking spaces t 1i 、Users reach i Straight-line driving time per parking space t 2i , and the user reaches i Parking time for each parking space t 3i .

[0114] In this embodiment, considering that most of the curves in the parking lot are right-angle curves, the curve driving time is t 1i The calculation can be done using the following formula:

[0115] ,

[0116] In the above formula, s 1irepresents the turning distance; v1 represents the maximum safe speed of the vehicle when driving on a curve in the parking lot; R represents the turning radius of the vehicle; Indicates the number of bends w i The calculated curve rotation angle; when w i =1, then =π / 2, when w i =2, then =π, …, and so on.

[0117] by Figure 5 Take a small parking lot that includes lateral parking spaces and vertical parking spaces as an example. If the user selects parking space No. 1 in the figure, there is no need to go through a bend. If the user selects parking space No. 2 in the figure, two bends need to be passed.

[0118] Specifically, the straight-line driving distance includes the speed-changing driving distance when entering or leaving a curve, and the constant-speed driving distance on the straight line of the curve. The curves in the parking lot are mostly right-angle curves, so the time taken for each speed-changing driving distance is roughly the same. Therefore, the straight-line driving time in this embodiment is t 2i The calculation formula is as follows:

[0119] ,

[0120] In the above formula, s 2i Indicates the total distance traveled in a straight line; s 0 Indicates the speed change driving distance before and after each curve; v 2 Indicates the uniform driving speed on a straight road section; s 0 Indicates the speed change time before and after each curve.

[0121] Likewise Figure 5 For example, if the user selects parking space No. 1 in the figure, the journey only includes a section of uniform speed driving. If the user selects parking space No. 2 in the figure, two sections of variable speed driving are required.

[0122] In the solution of this embodiment, it is assumed that each user has a similar driving skill level, and the expected time for the user to enter the parking space is only related to the type of parking space. The entrance time of the same type of parking space adopts a fixed value, which corresponds to the average time taken by most users to enter the parking space. Therefore, the parking entry time t 3i The calculation formula is as follows:

[0123] ,

[0124] In the above formula, Indicates the parking space type. =1, 2, 3 correspond to lateral parking spaces, diagonal parking spaces, and vertical parking spaces respectively; t 1 , t 2 and t 3 They represent the average parking time for lateral parking spaces, diagonal parking spaces, and vertical parking spaces. Figure 6 .

[0125] (3) According to the expected time T i The candidate parking spaces are sorted from short to long, and the top-ranked parking spaces are returned to the user as recommended parking spaces.

[0126] In actual application, the number of parking spaces in each parking lot is limited. In the process of solving the decision model, we can first traverse each free parking space, then calculate the expected time for each free parking space in turn, and finally generate a recommended parking space. The expected time for each free parking space can be calculated in real time, or it can be calculated in advance by the server, and then the relevant data can be directly called in the process of solving the decision model.

[0127] S5: Based on the current location of the user included in the request information, generate navigation information to the ideal parking space selected by the user, and return it to the user.

[0128] Example 2

[0129] This embodiment provides a parking guidance system in a parking lot, which adopts the planning method of the optimal path for autonomous parking in the parking lot as in Embodiment 1, so as to provide parking space recommendation and navigation services to the user when the user makes a parking request.

[0130] like Figure 7 As shown, the parking guidance system in a parking lot provided by the present invention includes: a data acquisition unit, a parking space state prediction network, a parking space dynamic release unit, a parking time calculation unit, a parking space recommendation unit, and a navigation unit.

[0131] The data collection unit is used to obtain the map information of the parking space distribution in each parking lot, and collect the parking space occupancy status and vehicle flow information in and out of the parking lot at each moment in real time. G、 Parking space status matrix C t and traffic flow feature vector S t .

[0132] The parking space status prediction network is trained by the T-GCN temporal graph convolutional network. The parking space status prediction network is used to predict the current parking lot based on the input graph. G And the parking space status matrix C t and traffic flow feature vector S t The historical sequence data is used to output the predicted parking space status matrix at the future moment. C t The sequence data.

[0133] The parking space dynamic release unit is used to obtain the future parking space status matrix output by the parking space status prediction network in real time. C t The available parking spaces in the parking lot are dynamically updated.

[0134] The parking time calculation unit is used to calculate the time required for the three stages of driving, namely, curve, straight line and parking, starting from the parking lot entrance and ending at each candidate parking space. t 1i , t 2i , t 3i , add the three together to get the expected time for each parking space T i .

[0135] The parking space recommendation unit is used to calculate the occupancy status of each parking space in the parking lot. c i is the decision coefficient, and each available free parking space is selected according to the expected time T i Sort in ascending order, generate a corresponding candidate list and send it to the user.

[0136] The navigation unit is used to provide navigation services to the user based on the user's current location and the parking space selected from the candidate list, and guide the user to complete parking.

[0137] Example 3

[0138] This embodiment provides a parking guidance device in a parking lot, which includes a memory, a processor, and a computer program stored in the memory and running in the processor. When the processor executes the computer program, the aforementioned parking lot autonomous parking optimal path planning method is adopted to provide parking space recommendation and navigation services to the user when the user issues a parking request.

[0139] The parking guidance device provided in this embodiment is essentially a computer device for implementing the solutions in embodiments 1 and 2. The computer device can be an embedded device installed in each vehicle. The embedded device can communicate and interact with the server and other devices to obtain the required data. And the data is processed by independent computing or edge computing. The computer device can also be a back-end device independent of the vehicle, and then complete the data processing process when the user makes a request, and send parking space recommendation information and navigation information to the data terminal in each vehicle according to the data processing results.

[0140] In this embodiment, the computer device may also be a smart terminal, tablet computer, laptop computer, desktop computer, rack server, blade server, tower server or cabinet server (including an independent server or a server cluster composed of multiple servers) that can execute programs.

[0141] The computer device indicated in this embodiment includes at least but is not limited to: a memory and a processor that can be connected to each other through a system bus. Among them, the memory (i.e., a readable storage medium) includes a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory can be an internal storage unit of a computer device, such as a hard disk or a memory of the computer device. In other embodiments, the memory can also be an external storage device of a computer device, such as a plug-in hard disk equipped on the computer device, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory can also include both the internal storage unit of the computer device and its external storage device. In this embodiment, the memory is generally used to store an operating system and various application software installed on the computer device. In addition, the memory can also be used to temporarily store various types of data that have been output or are to be output.

[0142] The processor may be a central processing unit (CPU), a graphics processing unit (GPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor is generally used to control the overall operation of a computer device. In this embodiment, the processor is used to run program codes stored in a memory or process data.

[0143] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for planning an optimal path for autonomous parking in a parking lot, which is used to provide assistance to the user's parking behavior decision-making process by combining the prediction results of the parking space occupancy status in the parking lot and the spatial distribution of the parking spaces, and is characterized in that: It includes the following steps: S1: Obtain a historical status data set of a designated parking lot, wherein the feature data in the historical status data set includes: a map information representing the distribution of parking spaces in the parking lot; G , representing the number of parking spaces in the parking lot t The parking space state matrix of the occupancy state at the time C t , and characterize parking lots in t Traffic flow feature vector of traffic flow information at the moment S t ; S2: Using the historical status dataset of each parking lot, a T-GCN temporal graph convolutional network consisting of a GCN layer and a GRU layer is trained. The trained network model is used as a parking space status prediction network for predicting the occupancy status of each parking space in the parking lot at a future time. S3: The vehicle reaches the designated parking space from the current position i Expected time T i To optimize the goal, a parking space selection decision model is established with the vacancy status of each parking space as a constraint; S4: When any user makes a parking lot planning request, the recommended parking spaces are generated and returned to the user through the following steps: (1) Inputting the historical status data set of each parking lot in the request area into the parking space status prediction network, and outputting a parking space status matrix sequence representing the occupancy status of each parking space at a future moment; (2) Combined with the dynamically updated parking space status matrix of the parking lot at future times, the decision model for parking space selection is solved to obtain the expected time for the user to reach each vacant candidate parking space. T i ; (3) According to the expected time T i Sort the candidate parking spaces from short to long, and return the top-ranked parking spaces to the user as recommended parking spaces; S5: Based on the current location of the user included in the request information, generate navigation information to the ideal parking space selected by the user, and return it to the user.

2. The method for planning the optimal path for autonomous parking in a parking lot according to claim 1, characterized in that: In step S1, a map representing the map information inside the parking lot G The data format is: G =( V , E ), where is the set of nodes corresponding to the parking spaces in the parking lot, V={ v 1 , v 2 , …, v i , …, v n }, v i Indicates the number of i parking spaces; E Represents a set of paths between parking spaces in a parking lot; The parking space state matrix C t Any element in represents the corresponding parking space in the parking lot v i exist t Occupancy status at the moment; The traffic feature vector S t The data format is: S t =( d t , q t , v t , p t , U t ) T ;in, d t Indicates the current day of the week; d t ∈{1,2,…,7}; q t Indicates the current parking lot traffic volume; v t Indicates the vehicle flow speed in the parking lot at the current moment; p t Indicates the location information of the vacant parking spaces in the parking lot at the current moment; U t Indicates the time required for the vehicle to leave the parking lot at the current moment.

3. The method for planning the optimal path for autonomous parking in a parking lot according to claim 2, characterized in that: In step S2, the T-GCN temporal graph convolutional network includes an input layer, a GCN layer, a GRU layer, and an output layer; The input layer is used to C t and S t A simple check of the sequence data was performed and compared with the G Input into the GCN layer together; The GCN layer learns complex topological structures through graph convolutional networks, captures spatial features between nodes in the graph G, and stacks multiple convolutional layers to obtain spatial correlation; The GRU layer is used to obtain the temporal features C t and S t The time series is input into the gated recurrent unit model, and the dynamic changes are obtained through information transmission between units to obtain the time characteristics of the vehicle flow information; the output layer is used to predict the parking space state matrix at the future moment according to the captured spatial characteristics and time characteristics and output it.

4. The method for planning the optimal path for autonomous parking in a parking lot according to claim 3, characterized in that: The state equation of the GCN layer is as follows: ， In the above formula, Represents the matrix after preprocessing the self-connection matrix; is the self-connection matrix of A, which is used to connect the nodes represented by each parking space and satisfies: ; I N is the identity matrix, is an adjacency matrix, which represents the connection relationship between parking space nodes and satisfies: , the elements in the matrix is 0 or 1, indicating the existence or non-existence of an edge between nodes; yes The output of the layer reflects the feature information between parking spaces; It is the parking space feature information after passing through the GCN layer. is a parameter containing the layer; Represents the sigmoid function; represents the weight matrix from the input layer to the hidden layer, P is the length of the feature matrix, including parking space status features, traffic flow information features, and parking lot map features. H is the number of hidden units; represents the weight matrix from the hidden layer to the output layer; Represents the predicted length T The output represents the updated features of each parking space node; Re LU is the activation function.

5. The method for planning the optimal path for autonomous parking in a parking lot according to claim 4, characterized in that: The state equation of the GRU layer is as follows: ， In the above formula, h t is the hidden state of the current node; represents the candidate hidden state; h t-1 The hidden state passed from the previous node; z t represents the output of the update gate; y t is the output of the current hidden node; Represents the sigmoid function; W 0 represents the initial weight matrix; x t is the current output; r t Represents the output of the reset gate; tanh is the hyperbolic tangent activation function; W z , W r , They are all concatenated vector weight matrices, which are also the parameters that the GRU model needs to learn.

6. The method for planning the optimal path for autonomous parking in a parking lot according to claim 5, characterized in that: In step S3, the expected time T i Travel time through curves t 1i , Straight-line driving time t 2i and parking time t 3i The curve driving time is composed of three parts: t 1i The calculation formula is as follows: ， In the above formula, s 1i represents the turning distance; v1 represents the maximum safe speed of the vehicle when driving on a curve in the parking lot; R represents the turning radius of the vehicle; Indicates the number of bends The calculated curve rotation angle; when =1, then =π / 2, when =2, then =π, and so on.

7. The method for planning the optimal path for autonomous parking in a parking lot according to claim 6, characterized in that: The straight-line driving time t 2i The calculation formula is as follows: ， In the above formula, s 2i Indicates the total distance traveled in a straight line; s 0 Indicates the speed change driving distance before and after each curve; v 2 Indicates the uniform driving speed on a straight road section; t 0 means the speed change time before and after each curve.

8. The method for planning the optimal path for autonomous parking in a parking lot according to claim 7, characterized in that: The parking time t 3i The calculation formula is as follows: ， In the above formula, Indicates the parking space type. =1, 2, 3 correspond to lateral parking spaces, diagonal parking spaces, and vertical parking spaces respectively; t 1 , t 2 and t 3 Respectively represent the average parking time for lateral parking spaces, diagonal parking spaces and vertical parking spaces.

9. A parking guidance system in a parking lot, characterized in that: The method for planning the optimal path for autonomous parking in a parking lot as claimed in any one of claims 1 to 8 is adopted to provide parking space recommendation and navigation services to the user when the user makes a parking request; the guidance system includes: The data acquisition unit is used to obtain the map information of the parking space distribution in each parking lot and collect the parking space occupancy status and vehicle flow information in and out of the parking lot at each moment in real time; then generate a map based on the collected data G、 Parking space status matrix C t and traffic flow feature vector S t ; The parking space status prediction network is trained by the T-GCN temporal graph convolutional network; the parking space status prediction network is used to predict the current parking lot according to the input graph G And the parking space status matrix C t and traffic flow feature vector S t The historical sequence data is used to output the predicted parking space status matrix at the future moment. C t Sequence data of The parking space dynamic release unit is used to obtain the future parking space state matrix output by the parking space state prediction network in real time. C t The sequence data of the parking lot is collected and the available free parking spaces are dynamically updated; The parking time calculation unit is used to calculate the time required for the three stages of driving, namely, curve, straight line and parking, starting from the parking lot entrance and ending at each candidate parking space. t 1i , t 2i , t 3i , add the three together to get the expected time for each parking space T i ; Parking space recommendation unit, which is used to recommend the occupancy status of each parking space in the parking lot c i is the decision coefficient, and each available free parking space is selected according to the expected time T i Sort in ascending order, generate a corresponding candidate list and send it to the user; The navigation unit is used to provide navigation services to the user according to the user's current location and the parking space selected from the candidate list, and guide the user to complete parking.

10. A parking guidance device in a parking lot, comprising a memory, a processor, and a computer program stored in the memory and running in the processor, characterized in that: When the processor executes the computer program, the method for planning the optimal path for autonomous parking in a parking lot as described in any one of claims 1 to 8 is adopted to provide parking space recommendation and navigation services to the user when the user makes a parking request.

Images