Multi-dimensional prediction method and system for parking lot occupancy

Abstract

The application discloses a multi-dimensional prediction method and system for parking space occupancy of a parking lot. The method comprises the following steps: acquiring inter-parking lot straight-line distances between parking lots and parking space occupancy time series training sets of the parking lots; obtaining inter-parking lot relationship data between the parking lots according to the inter-parking lot straight-line distances; training a graph neural network according to the inter-parking lot relationship data and the parking space occupancy time series training sets, and obtaining parking space spatial features and parking space occupancy time series of each parking lot; training a long short-term memory neural network according to the parking space spatial features and the parking space occupancy time series of each parking lot, and obtaining time series features of parking space occupancy quantity of each parking lot; and inputting the time series features of parking space occupancy quantity of each parking lot into a full connection layer for nonlinear feature extraction, and obtaining parking space occupancy of each parking lot at a next interval time.

Description

Technical Field

[0001] This application relates to the field of smart parking technology, and in particular to a multi-dimensional prediction method and system for parking space occupancy. Background Technology

[0002] With the significant increase in car ownership in my country, urban parking services have flourished, leading to a substantial increase in the number of parking spaces. However, parking resources remain relatively scarce, especially in bustling urban areas. Therefore, a high-precision, high-fitting-degree parking space prediction method can better improve the utilization rate of parking space resources.

[0003] Traditional forecasting methods use linear forecasting, which have simple models but are weak in handling data volatility, resulting in poor stability of forecast results. Summary of the Invention

[0004] The purpose of this application is to address the technical problem of poor stability in prediction results caused by the weak ability of traditional prediction methods to handle data volatility. To achieve the above objective, this application provides a multi-dimensional prediction method and system for parking lot space occupancy.

[0005] This application provides a multi-dimensional prediction method for parking space occupancy, including:

[0006] Obtain the straight-line distance between parking lots and the time-series training set of parking space occupancy for each parking lot;

[0007] Based on the straight-line distance between each parking lot, obtain the parking lot relationship data between each parking lot;

[0008] Based on the data on the relationship between parking lots and the time series training set of parking space occupancy, a graph neural network is trained to obtain the parking lot spatial features and parking space occupancy time series sequence of each parking lot.

[0009] Based on the parking lot space characteristics and the parking space occupancy time sequence of each parking lot, a long short-term memory neural network is trained to obtain the time sequence characteristics of the number of parking spaces occupied in each parking lot.

[0010] The temporal features of the number of occupancy spaces in each parking lot are input into a fully connected layer for nonlinear feature extraction to obtain the number of occupancy spaces in each parking lot at the next interval.

[0011] In one embodiment, obtaining the inter-parking lot relationship data based on the straight-line distance between each parking lot includes:

[0012] Construct a two-dimensional distance matrix based on the straight-line distance between each of the parking lots;

[0013] Calculate the weight parameters between the various parking lots according to the weight calculation formula;

[0014] Based on the weight parameters, the inter-parking relationship data between the various parking lots is obtained;

[0015] The weight calculation formula is as follows:

[0016]

[0017] d ij The distance between the i-th and j-th parking lots in the two-dimensional distance matrix represents the straight-line distance between them, where σ and ε are formula parameters.

[0018] In one embodiment, training a graph neural network based on the inter-parking lot relationship data and the parking space occupancy time-series training set to obtain the parking lot spatial features and parking space occupancy time-series sequence of each parking lot includes:

[0019] Construct the graph neural network, input the parking lot relationship data and the parking space occupancy time series training set into the graph neural network, initialize the parking space occupancy time series training set with uniformly distributed weights, and obtain initialized parking space occupancy time series data;

[0020] Based on the berth occupancy time series training set and the relationship data between parking lots, obtain the berth occupancy time series data between adjacent parking lots;

[0021] Aggregated data is obtained based on the time-series data of parking space occupancy between adjacent parking lots, the data on the relationship between parking lots, and the number of parking lots;

[0022] Based on the aggregated data and the initialized berth occupancy time series data, the spliced ​​data is obtained;

[0023] The stitched data is input into the activation layer to obtain the parking space features and parking space occupancy time series sequence of each parking lot in the parking space occupancy time series training set.

[0024] In one embodiment, obtaining the straight-line distance between parking lots and the time-series training set of parking space occupancy for each parking lot includes:

[0025] Obtain parking record data from each parking lot, perform missing data supplementation, normalization, and data splitting on the parking record data to obtain the parking space occupancy time series training set and the parking space occupancy time series test set;

[0026] The parking space occupancy time-series training set and the parking space occupancy time-series test set are two-dimensional matrices of the number of parking spaces occupied in each parking lot, with time intervals as the time sequence.

[0027] In one embodiment, the method further includes:

[0028] Based on the berth occupancy time series test set, the berth occupancy prediction model formed by the graph neural network, the long short-term memory neural network, and the fully connected layer is tested and evaluated to obtain the tested berth occupancy prediction model.

[0029] This application provides a multi-dimensional prediction system for parking space occupancy, including:

[0030] The parking lot data acquisition module is used to acquire the straight-line distance between parking lots and the time-series training set of parking space occupancy of each parking lot;

[0031] The parking lot relationship acquisition module is used to obtain parking lot relationship data between the parking lots based on the straight-line distance between each parking lot;

[0032] The graph neural network module is used to train the graph neural network based on the inter-parking lot relationship data and the parking space occupancy time series training set to obtain the parking lot spatial features and parking space occupancy time series sequence of each parking lot.

[0033] The Long Short-Term Memory Neural Network module is used to train the Long Short-Term Memory Neural Network based on the parking lot space characteristics and the parking space occupancy time sequence of each parking lot, so as to obtain the time sequence characteristics of the number of parking spaces occupied in each parking lot.

[0034] The fully connected layer module is used to input the temporal features of the number of parking spaces occupied in each parking lot into the fully connected layer for nonlinear feature extraction, so as to obtain the number of parking spaces occupied in each parking lot at the next time interval.

[0035] In one embodiment, the inter-parking lot relationship acquisition module includes:

[0036] A two-dimensional distance matrix acquisition module is used to construct a two-dimensional distance matrix based on the straight-line distance between each of the parking lots;

[0037] The weight parameter acquisition module is used to calculate the weight parameters between the parking lots according to the weight calculation formula.

[0038] The relationship acquisition module is used to obtain the inter-parking relationship data between the various parking lots based on the weight parameters;

[0039] The weight calculation formula is as follows:

[0040]

[0041] d ijThe distance between the i-th and j-th parking lots in the two-dimensional distance matrix represents the straight-line distance between them, where σ and ε are formula parameters.

[0042] In one embodiment, the graph neural network module includes:

[0043] An initialization module is used to construct the graph neural network, input the inter-parking lot relationship data and the parking space occupancy time series training set into the graph neural network, perform uniform weight initialization on the parking space occupancy time series training set, and obtain initialized parking space occupancy time series data.

[0044] The adjacent parking lot data acquisition module is used to obtain the parking lot occupancy time series data between adjacent parking lots based on the parking lot occupancy time series training set and the parking lot relationship data.

[0045] The aggregated data acquisition module is used to obtain aggregated data based on the time-series data of parking space occupancy between adjacent parking lots, the relationship data between parking lots, and the number of parking lots;

[0046] The splicing data acquisition module is used to obtain splicing data based on the aggregated data and the initialized berth occupancy time series data;

[0047] The parking lot space feature acquisition module is used to input the stitched data into the activation layer to obtain the parking lot space features and parking space occupancy time series of each parking lot in the parking space occupancy time series training set.

[0048] In one embodiment, the parking lot data acquisition module includes:

[0049] The training and test set acquisition module is used to acquire parking record data of each parking lot, perform missing data supplementation, normalization and data splitting on the parking record data, and obtain the parking space occupancy time series training set and the parking space occupancy time series test set.

[0050] The parking space occupancy time-series training set and the parking space occupancy time-series test set are two-dimensional matrices of the number of parking spaces occupied in each parking lot, with time intervals as the time sequence.

[0051] In one embodiment, the system further includes:

[0052] The testing module is used to test and evaluate the berth occupancy prediction model formed by the graph neural network, the long short-term memory neural network, and the fully connected layer based on the berth occupancy time series test set, and obtain the tested berth occupancy prediction model.

[0053] The aforementioned multi-dimensional prediction method and system for parking lot occupancy combines graph neural networks and long short-term memory neural networks. Based on the straight-line distance between parking lots, the graph neural network constructs the network topology between each parking lot, enabling the extraction of parking lot spatial features from the parking lot occupancy time-series training set. Furthermore, based on the parking lot occupancy time-series sequences in the training set, the long short-term memory neural network learns the temporal variation patterns, extracting the temporal features of the parking lot occupancy quantity for each parking lot in the training set. The output of the fully connected layer then yields the parking lot occupancy quantity for each parking lot at the next interval.

[0054] The multi-dimensional prediction method and system for parking space occupancy provided in this application fully consider the temporal and spatial characteristics of each parking lot in the parking space occupancy time-series training set, take into account the spatial influence of other parking lots around each parking lot, reduce the model parameters of the long short-term memory neural network, reduce the training difficulty, and improve the prediction accuracy of parking space occupancy of each parking lot at the next interval. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the steps of the multi-dimensional prediction method for parking space occupancy provided in this application.

[0056] Figure 2 This is a schematic diagram of the structure of the multi-dimensional prediction system for parking space occupancy provided in this application;

[0057] Figure 3 It is an evaluation indicator of the multi-dimensional prediction model of parking space occupancy provided in this application;

[0058] Figure 4 This application provides a multi-dimensional prediction system for parking space occupancy. Detailed Implementation

[0059] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0060] Please see Figure 1 This application provides a multi-dimensional prediction method for parking space occupancy, including:

[0061] S10, obtain the straight-line distance between parking lots and the time series training set of parking space occupancy in each parking lot;

[0062] S20: Based on the straight-line distance between each parking lot, obtain the data on the relationship between each parking lot;

[0063] S30. Based on the data on the relationship between parking lots and the time series training set of parking space occupancy, a graph neural network is trained to obtain the parking lot spatial characteristics and parking space occupancy time series sequence of each parking lot.

[0064] S40, based on the parking lot space characteristics and parking space occupancy time sequence of each parking lot, train the long short-term memory neural network to obtain the time sequence characteristics of the number of parking spaces occupied in each parking lot;

[0065] S50 inputs the temporal features of the number of parking spaces occupied in each parking lot into the fully connected layer for nonlinear feature extraction, and obtains the number of parking spaces occupied in each parking lot at the next interval.

[0066] In this embodiment, each parking lot can be multiple parking lots within the same city. There is a straight-line distance between any two parking lots. Each parking lot corresponds to a number of occupied parking spaces at each time interval. For example, at 10:30 on September 1, 2022, parking lot A has 75 occupied parking spaces, parking lot B has 80 occupied parking spaces, parking lot C has 45 occupied parking spaces, and so on. Using the time interval as the time sequence, the number of occupied parking spaces for each parking lot is obtained, and all data is divided to form a parking space occupancy time-series training set and a parking space occupancy time-series test set.

[0067] There is a straight-line distance between any two parking lots. This straight-line distance establishes connections between the parking lots, representing their spatial characteristics. The straight-line distances between parking lots yield data on the relationships between them. This data, along with a time-series training set of parking space occupancy, is used to train graph neural networks and long short-term memory neural networks.

[0068] Graph Neural Network (GNN) layers, Long Short-Term Memory (LSTM) layers, and fully connected layers can extract spatial features, temporal features, and summarize all features, respectively. The data on relationships between parking lots is input into the GNN to construct the network topology between parking lots, establishing distance relationships between each parking lot. The straight-line distances between parking lots connect them, thus establishing a relationship between the number of occupant parking spaces in each parking lot. The data on relationships between parking lots is used as parameters for the GNN layers, which extract spatial features between parking lots from the time-series training set of parking space occupancy, outputting the spatial features of each parking lot and the time-series sequence of parking space occupancy.

[0069] A Long Short-Term Memory (LSTM) neural network is connected after the output of the graph neural network. The LTM network extracts the temporal features from the parking space occupancy time series to obtain the temporal features of the number of parking spaces occupied in each parking lot. The output of the graph neural network serves as the input to the LTM network, learning the temporal variation patterns of each parking space occupancy time series in the training set, and extracting the temporal features of the number of parking spaces occupied in each parking lot. Through a fully connected layer and nonlinear transformation, the output feature information of the LTM network is summarized to output the number of parking spaces occupied in each parking lot at the next interval, thus achieving the prediction of the number of parking spaces occupied in the parking lot.

[0070] The multi-dimensional prediction method for parking space occupancy provided in this application combines graph neural networks and long short-term memory neural networks. Based on the straight-line distance between parking lots, the graph neural network is used to construct the network topology between each parking lot, thereby extracting the parking lot spatial features of each parking lot in the parking space occupancy time-series training set. Furthermore, based on the parking space occupancy time-series sequences of each parking space in the parking space occupancy time-series training set, the long short-term memory neural network is used to learn the time change pattern, thereby extracting the time-series features of the number of parking spaces occupied in each parking lot in the parking space occupancy time-series training set. After passing through the output of the fully connected layer, the parking space occupancy of each parking lot at the next interval is obtained.

[0071] The multi-dimensional prediction method for parking space occupancy provided in this application fully considers the temporal and spatial features of each parking lot in the parking space occupancy time-series training set, takes into account the spatial influence of other parking lots around each parking lot, reduces the model parameters of the long short-term memory neural network, lowers the training difficulty, and improves the prediction accuracy of parking space occupancy of each parking lot at the next interval.

[0072] In one embodiment, when training the graph neural network, long short-term memory neural network, and fully connected layers, the optimizer uses the RMSprop algorithm, the learning rate is set to 0.001, and the loss function is the mean squared error function.

[0073] In one embodiment, S10, the straight-line distance between parking lots and the time-series training set of parking space occupancy for each parking lot are obtained, including:

[0074] S110: Obtain parking record data from each parking lot, perform missing data supplementation, normalization, and data splitting on the parking record data to obtain the parking space occupancy time series training set and the parking space occupancy time series test set.

[0075] The berth occupancy time series training set and berth occupancy time series test set are two-dimensional matrices of the number of berths occupied in each parking lot, with the time interval as the time sequence.

[0076] In this embodiment, the parking record data can be data from hundreds of parking lots within a city. For one embodiment, please refer to... Figure 2 The example shown is as follows:

[0077] At 10:30 on September 1, 2022, parking lot A had 75 occupied spaces, parking lot B had 80 occupied spaces, parking lot C had 0 occupied spaces, and so on.

[0078] At 11:00 on September 1, 2022, the number of occupied parking spaces in Parking Lot A was 74, the number of occupied parking spaces in Parking Lot B was 78, and the number of occupied parking spaces in Parking Lot C was 0 at 10:30 on September 1, 2022, and so on.

[0079] At 11:30 on September 1, 2022, the number of occupied parking spaces in Parking Lot A was 78, the number of occupied parking spaces in Parking Lot B was 81, and the number of occupied parking spaces in Parking Lot C was 0 at 10:30 on September 1, 2022, and so on.

[0080] The interval can be set to 10 minutes, 20 minutes, 30 minutes, or 40 minutes, etc., depending on actual needs. In one embodiment, the parking record data of each parking lot is collected during weekdays, which can eliminate the influence of atypical data.

[0081] Missing parking record data is supplemented using linear imputation, and normalized using either Z-Score or linear normalization methods. The data is then further split into a parking space occupancy time-series training set and a parking space occupancy time-series test set. If parking record data is missing due to equipment or system issues, parking record data from parking lots in neighboring cities can be used to supplement the missing data.

[0082] After performing missing data supplementation, normalization, and data splitting on the parking record data in sequence, it is converted into an m*n two-dimensional matrix, where n represents the number of parking lots (n can be set to 328) and m represents the number of time series data (m can be set to 2*24*365=17520), which can be understood as the number of time intervals.

[0083] In one embodiment, S20, based on the straight-line distance between each parking lot, obtains the inter-parking lot relationship data, including:

[0084] S210, construct a two-dimensional distance matrix based on the straight-line distance between each parking lot;

[0085] S220, calculate the weight parameters between each parking lot according to the weight calculation formula;

[0086] S230, based on the weight parameters, obtains the data on the relationship between parking lots;

[0087] The formula for calculating the weights is as follows:

[0088]

[0089] d ij Let σ and ε represent the straight-line distance between the i-th and j-th parking lots in the two-dimensional distance matrix, where σ and ε are formula parameters.

[0090] In this embodiment, the straight-line distance between each parking lot is converted into an n*n two-dimensional matrix. n represents the number of parking spaces, and the element (i, j) in row i and column j represents the distance from the i-th parking lot to the j-th parking lot. The two-dimensional distance matrix reveals the distance relationships between the parking lots. Weights ω are calculated using the weight calculation formula provided in this embodiment. ij The values ​​of σ and ε can be 0.1 and 0.5, respectively. ω ij This data represents the strength of the relationships between parking lots. A value of 0 indicates no relationship between the two parking lots, while a positive value indicates a stronger relationship and mutual influence between them. Multiple weight parameters are calculated using a weighting formula, forming the data on the relationships between the parking lots. Therefore, this data can characterize the spatial characteristics of each parking lot.

[0091] In one embodiment, S30, a graph neural network is trained based on the inter-parking lot relationship data and the parking space occupancy time series training set to obtain the parking lot spatial features and parking space occupancy time series sequence for each parking lot, including:

[0092] S310, Construct a graph neural network, input the inter-parking lot relationship data and the berth occupancy time series training set into the graph neural network, perform uniform weight initialization on the berth occupancy time series training set, and obtain the initialized berth occupancy time series data;

[0093] S320: Based on the time series training set of parking space occupancy and the relationship data between parking lots, obtain the time series data of parking space occupancy between adjacent parking lots;

[0094] S330: Aggregated data is obtained based on the time series data of parking space occupancy between adjacent parking lots, the relationship data between parking lots, and the number of parking lots.

[0095] S340, obtain spliced ​​data based on aggregated data and initial berth occupancy time series data;

[0096] S350 inputs the stitched data into the activation layer to obtain the parking space characteristics and parking space occupancy time series sequence of each parking lot in the parking space occupancy time series training set.

[0097] In this embodiment, the uniform distribution weight initialization can be Glorot uniform distribution weights, which are used to initialize the parking space occupancy time-series training set to obtain initialized parking space occupancy time-series data. The parking space occupancy time-series data between adjacent parking lots can also be understood as slices of parking space occupancy time-series data between adjacent parking lots. Adjacent parking lots can be understood as the two parking lots with the largest corresponding weight coefficient values. The aggregated data integrates the parking space occupancy time-series data between adjacent parking lots, the relationship data between parking lots, and the number of parking lots. After concatenating the aggregated data incorporating the relationship data between parking lots with the initialized parking space occupancy time-series data using a data tensor, the concatenated data is obtained. After passing through an activation layer, the concatenated data outputs the parking lot spatial features and parking space occupancy time-series sequence of each parking lot in the parking space occupancy time-series training set, thus achieving filtering of the relationships between various parking lots in the parking space occupancy time-series training set.

[0098] In one embodiment, the multi-dimensional prediction method for parking space occupancy further includes:

[0099] S60. Based on the berth occupancy time series test set, the berth occupancy prediction model formed by graph neural network, long short-term memory neural network and fully connected layer is tested and evaluated to obtain the tested berth occupancy prediction model.

[0100] In this embodiment, the parking space occupancy prediction model includes a graph neural network, a long short-term memory neural network, and a fully connected layer. These layers extract both temporal and spatial features from the parking space occupancy time-series training set, and the fully connected layer then aggregates and extracts these features to predict the parking space occupancy of each parking lot at the next time interval. The model is then tested and evaluated using a parking space occupancy time-series test set to validate its performance. The mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) can be used as evaluation functions for the prediction results during validation using the test set.

[0101] Please see Figure 3As shown in the table, the parking space occupancy prediction model based on graph neural networks and long short-term memory neural networks significantly outperforms other prediction models. The model predicts the parking space occupancy of each parking lot at the next time interval in 319 seconds, far shorter than traditional prediction methods. On larger datasets, the training performance will be further improved. In the multi-dimensional parking space occupancy prediction method provided in this application, the GNN-LSTM model has only 32% of the parameters of the LSTM model, allowing for faster and easier convergence during training.

[0102] Please see Figure 4 This application provides a multi-dimensional prediction system 100 for parking space occupancy. The multi-dimensional prediction system 100 includes a parking lot data acquisition module 10, a parking lot relationship acquisition module 20, a graph neural network module 30, a long short-term memory neural network module 40, and a fully connected layer module 50. The parking lot data acquisition module 10 is used to acquire the straight-line distance between parking lots and the time-series training set of parking space occupancy for each parking lot. The parking lot relationship acquisition module 20 is used to obtain the inter-parking lot relationship data based on the straight-line distance between parking lots.

[0103] Graph Neural Network Module 30 is used to train the graph neural network based on the inter-parking lot relationship data and the parking space occupancy time-series training set to obtain the parking lot spatial features and parking space occupancy time-series sequence for each parking lot. Long Short-Term Memory Neural Network Module 40 is used to train the long short-term memory neural network based on the parking lot spatial features and parking space occupancy time-series sequence for each parking lot to obtain the time-series features of the number of parking spaces occupied for each parking lot. Fully Connected Layer Module 50 is used to input the time-series features of the number of parking spaces occupied for each parking lot into the fully connected layer for nonlinear feature extraction to obtain the parking space occupancy amount for each parking lot at the next interval.

[0104] In this embodiment, the description of the parking lot data acquisition module 10 can be found in the description of S10 in the above embodiment. The description of the parking lot relationship acquisition module 20 can be found in the description of S20 in the above embodiment. The description of the graph neural network module 30 can be found in the description of S30 in the above embodiment. The description of the long short-term memory neural network module 40 can be found in the description of S40 in the above embodiment. The description of the fully connected layer module 50 can be found in the description of S50 in the above embodiment.

[0105] In one embodiment, the parking lot relationship acquisition module 20 includes a two-dimensional distance matrix acquisition module, a weight parameter acquisition module, and a relationship acquisition module. The two-dimensional distance matrix acquisition module constructs a two-dimensional distance matrix based on the straight-line distance between each parking lot. The weight parameter acquisition module calculates the weight parameters between each parking lot according to a weight calculation formula. The relationship acquisition module obtains the parking lot relationship data between each parking lot based on the weight parameters.

[0106] The formula for calculating the weights is as follows:

[0107]

[0108] d ij Let σ and ε represent the straight-line distance between the i-th and j-th parking lots in the two-dimensional distance matrix, where σ and ε are formula parameters.

[0109] In this embodiment, the description of the two-dimensional distance matrix acquisition module can be found in the description of S210 in the above embodiment. The description of the weight parameter acquisition module can be found in the description of S220 in the above embodiment. The description of the relationship acquisition module can be found in the description of S230 in the above embodiment.

[0110] In one embodiment, the graph neural network module 30 includes an initialization module, an adjacent parking lot data acquisition module, an aggregated data acquisition module, a stitched data acquisition module, and a parking lot spatial feature acquisition module. The initialization module constructs the graph neural network by inputting the parking lot relationship data and the parking space occupancy time-series training set into the graph neural network, performing uniform weight initialization on the parking space occupancy time-series training set to obtain initialized parking space occupancy time-series data. The adjacent parking lot data acquisition module obtains adjacent parking lot occupancy time-series data based on the parking space occupancy time-series training set and the parking lot relationship data. The aggregated data acquisition module obtains aggregated data based on the adjacent parking lot occupancy time-series data, the parking lot relationship data, and the number of parking lots. The stitched data acquisition module obtains stitched data based on the aggregated data and the initialized parking space occupancy time-series data. The parking lot spatial feature acquisition module inputs the stitched data into the activation layer to obtain the parking lot spatial features and parking space occupancy time-series sequence of each parking lot in the parking space occupancy time-series training set.

[0111] In this embodiment, the description of the initialization module can be found in the description of S310 in the above embodiment. The description of the data acquisition module between adjacent parking lots can be found in the description of S320 in the above embodiment. The description of the aggregated data acquisition module can be found in the description of S330 in the above embodiment. The description of the stitched data acquisition module can be found in the description of S340 in the above embodiment. The description of the parking lot spatial feature acquisition module can be found in the description of S350 in the above embodiment.

[0112] In one embodiment, the parking lot data acquisition module 10 includes a training and test set acquisition module. This module acquires parking record data from each parking lot, performs missing data supplementation, normalization, and data splitting on the parking record data, and obtains a parking space occupancy time-series training set and a parking space occupancy time-series test set. The parking space occupancy time-series training set and test set are two-dimensional matrices showing the number of parking spaces occupied in each parking lot at time intervals.

[0113] In this embodiment, the description of the training and test set acquisition module can be found in the description of S110 in the above embodiment.

[0114] In one embodiment, the multi-dimensional prediction system 100 for parking space occupancy further includes a testing module. The testing module is used to test and evaluate the parking space occupancy prediction model formed by the graph neural network, long short-term memory neural network, and fully connected layers based on a parking space occupancy time-series test set, thereby obtaining a tested parking space occupancy prediction model.

[0115] In this embodiment, the relevant description of the test module can be referred to the description of S60 in the above embodiment.

[0116] In the various embodiments described above, the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process may be rearranged without departing from the scope of this disclosure. The appended method claims provide elements of various steps in an exemplary order and are not intended to limit the scope to a specific order or hierarchy.

[0117] Those skilled in the art will also understand that the various illustrative logical blocks, modules, and steps listed in the embodiments of this application can be implemented by electronic hardware, computer software, or a combination of both. To clearly demonstrate the interchangeability of hardware and software, the functions of the various illustrative components, modules, and steps described above have been generally described. Whether such functionality is implemented through hardware or software depends on the specific application and the overall system design requirements. Those skilled in the art can implement the described functions using various methods for each specific application, but such implementation should not be construed as exceeding the scope of protection of the embodiments of this application.

[0118] The various illustrative logic blocks or modules described in the embodiments of this application can be implemented or operate the described functions using a general-purpose processor, digital signal processor, application-specific integrated circuit (ASIC), field-programmable gate array or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof. The general-purpose processor can be a microprocessor; alternatively, it can be any conventional processor, controller, microcontroller, or state machine. The processor can also be implemented using a combination of computing devices, such as a digital signal processor and a microprocessor, multiple microprocessors, one or more microprocessors combined with a digital signal processor core, or any other similar configuration.

[0119] The steps of the methods or algorithms described in the embodiments of this application can be directly embedded in hardware, a software module executed by a processor, or a combination of both. The software module can be stored in RAM, flash memory, ROM, EPROM, EEPROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium in the art. Exemplarily, the storage medium can be connected to the processor so that the processor can read information from and write information to the storage medium. Optionally, the storage medium can also be integrated into the processor. The processor and storage medium can be housed in an ASIC, which can be housed in a user terminal. Optionally, the processor and storage medium can also be housed in different components of the user terminal.

[0120] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for multi-dimensional prediction of parking lot occupancy, characterized in that, The method comprises the following steps: obtaining the inter-parking lot straight-line distances between each parking lot and the parking space occupancy time sequence training set of each parking lot; obtaining the inter-parking lot relationship data between each parking lot according to each inter-parking lot straight-line distance; training a graph neural network according to the inter-parking lot relationship data and the parking space occupancy time sequence training set, to obtain the parking space spatial features and the parking space occupancy time sequence of each parking lot; training a long short-term memory neural network according to the parking space spatial features and the parking space occupancy time sequence of each parking lot, to obtain the time sequence features of the parking space occupancy quantity of each parking lot; inputting the time sequence features of the parking space occupancy quantity of each parking lot into a fully connected layer for nonlinear feature extraction, to obtain the parking space occupancy quantity of each parking lot at the next interval time; The method comprises the following steps: constructing a two-dimensional distance matrix according to each inter-parking lot straight-line distance; calculating the weight parameters between each parking lot according to a weight calculation formula; Based on the weight parameters, the inter-parking lot relationship data is obtained; wherein, the weight calculation formula is: Represents the second position in the two-dimensional distance matrix. i The parking lot and the first j The straight-line distance between the parking lots. These are formula parameters. 2.The method of claim 1, wherein, The method comprises the following steps: constructing the graph neural network, inputting the inter-parking lot relationship data and the parking space occupancy time sequence training set into the graph neural network, initializing the parking space occupancy time sequence data with uniform distribution weight according to the parking space occupancy time sequence training set, to obtain the initialized parking space occupancy time sequence data; obtaining the adjacent inter-parking lot parking space occupancy time sequence data according to the parking space occupancy time sequence training set and the inter-parking lot relationship data; obtaining the aggregation data according to the adjacent inter-parking lot parking space occupancy time sequence data, the inter-parking lot relationship data and the number of parking lots; obtaining the splicing data according to the aggregation data and the initialized parking space occupancy time sequence data; inputting the splicing data into an activation layer to obtain the parking space spatial features and the parking space occupancy time sequence of each parking lot in the parking space occupancy time sequence training set. 3.The method of claim 1, wherein, The method comprises the following steps: obtaining the parking record data of each parking lot, supplementing the missing data, normalizing and performing data splitting processing on the parking record data, to obtain the parking space occupancy time sequence training set and the parking space occupancy time sequence test set; The parking space occupancy time sequence training set and the parking space occupancy time sequence test set are two-dimensional matrices of the parking space occupancy quantity of each parking lot at interval time.

4. The method of claim 3, wherein, The method further comprises the following steps: testing and evaluating the parking space occupancy quantity prediction model formed by the graph neural network, the long short-term memory neural network and the fully connected layer according to the parking space occupancy time sequence test set, to obtain the tested parking space occupancy quantity prediction model.

5. A multi-dimensional prediction system for parking lot occupancy, characterized in that, The method comprises the following steps: a parking lot data acquisition module for obtaining the inter-parking lot straight-line distances between each parking lot and the parking space occupancy time sequence training set of each parking lot; The inter-parking lot relationship acquisition module is configured to acquire inter-parking lot relationship data between the parking lots according to the inter-parking lot straight-line distances. The graph neural network module is configured to train a graph neural network according to the inter-parking lot relationship data and the parking space occupancy time sequence training set, and to acquire parking lot space features and parking space occupancy time sequence of each parking lot. The long short-term memory neural network module is configured to train a long short-term memory neural network according to the parking lot space features and the parking space occupancy time sequence of each parking lot, and to acquire time sequence features of the number of parking spaces occupied by each parking lot. The full connection layer module is configured to input the time sequence features of the number of parking spaces occupied by each parking lot into a full connection layer for nonlinear feature extraction, and to acquire the number of parking spaces occupied by each parking lot at a next interval time. The inter-parking lot relationship acquisition module includes: The two-dimensional distance matrix acquisition module is configured to construct a two-dimensional distance matrix according to the inter-parking lot straight-line distances. The weight parameter acquisition module is configured to calculate weight parameters between the parking lots according to a weight calculation formula. The relationship acquisition module is configured to acquire the inter-parking lot relationship data between the parking lots according to the weight parameters. wherein the weight calculation formula is: denotes the inter-parking lot straight line distance between the i-th parking lot and the j-th parking lot in the two-dimensional distance matrix, i j denotes the inter-parking lot straight line distance between the i-th parking lot and the j-th parking lot in the two-dimensional distance matrix, is a formula parameter.​ 6. The multi-dimensional prediction system for parking facility occupancy of claim 5, wherein, The graph neural network module includes: The initialization module is configured to construct the graph neural network, to input the inter-parking lot relationship data and the parking space occupancy time sequence training set into the graph neural network, to perform uniform distribution weight initialization on the parking space occupancy time sequence training set, and to acquire initialized parking space occupancy time sequence data. The adjacent inter-parking lot data acquisition module is configured to acquire adjacent inter-parking lot parking space occupancy time sequence data according to the parking space occupancy time sequence training set and the inter-parking lot relationship data. The aggregated data acquisition module is configured to acquire aggregated data according to the adjacent inter-parking lot parking space occupancy time sequence data, the inter-parking lot relationship data, and the number of parking lots. The spliced data acquisition module is configured to acquire spliced data according to the aggregated data and the initialized parking space occupancy time sequence data. The parking lot space feature acquisition module is configured to input the spliced data into an activation layer, and to acquire parking lot space features and parking space occupancy time sequence of each parking lot in the parking space occupancy time sequence training set.

7. The multi-dimensional prediction system for parking facility occupancy of claim 5, wherein, The parking lot data acquisition module includes: The training and test set acquisition module is configured to acquire parking record data of the parking lots, to perform missing data supplement, normalization, and data split processing on the parking record data, and to acquire the parking space occupancy time sequence training set and the parking space occupancy time sequence test set. The parking space occupancy time sequence training set and the parking space occupancy time sequence test set are two-dimensional matrices of the number of parking spaces occupied by the parking lots at interval time.

8. The multi-dimensional prediction system for parking facility occupancy according to claim 7, wherein, The system further includes: The test module is configured to test and evaluate a parking space occupancy amount prediction model formed by the graph neural network, the long short-term memory neural network, and the full connection layer according to the parking space occupancy time sequence test set, and to acquire the tested parking space occupancy amount prediction model.

Images