Parking lot remaining space prediction method, equipment and parking lot management system

Abstract

The present application provides a parking lot remaining space prediction method, equipment and parking lot management system. In an example of the present application, the parking lot remaining space prediction method includes: clustering the parking lots to be predicted based on the clustering reference attributes of the parking lots to obtain k types of parking lots to be predicted; wherein k≥2; dividing each type of parking lot to be predicted into time periods to obtain the optimal time period division method for each type of parking lot to be predicted; training the remaining space prediction model for each target time period of each type of parking lot to be predicted based on the historical remaining space data of the parking lot to be predicted, and obtaining the remaining space prediction model corresponding to each target time period of each type of parking lot to be predicted. This method can improve the accuracy of parking lot remaining space prediction.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a parking lot remaining space prediction method, equipment and a parking lot management system. Background Art

[0002] With the rapid development of social economy, the number of vehicles is increasing. While vehicles bring convenience to people's work and life, the problem of parking difficulties has gradually become a problem that people need to face in their daily work and life.

[0003] In order to enable car owners to find parking spaces more efficiently and reduce the time and effort required to find parking spaces, parking space prediction has gradually become a hot research direction.

[0004] How to more accurately predict remaining parking spaces has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of this, the present application provides a parking lot remaining space prediction method, device and parking lot management system.

[0006] Specifically, the present application is implemented through the following technical solutions:

[0007] According to a first aspect of an embodiment of the present application, a method for predicting remaining parking spaces is provided, comprising:

[0008] According to the clustering reference attributes of the parking lot, the parking lot to be predicted is clustered to obtain k types of parking lots to be predicted; where k ≥ 2;

[0009] Divide each type of parking lot to be predicted into time periods, and obtain the optimal time period division method for each type of parking lot to be predicted;

[0010] According to the historical remaining space data of the parking lot to be predicted, the remaining space prediction model is trained for each target time period of each type of parking lot to be predicted, and the remaining space prediction models corresponding to each target time period of each type of parking lot to be predicted are obtained respectively; wherein, for any type of parking lot to be predicted, the target time period is the time period obtained by dividing according to the optimal time period division method of the parking lot to be predicted; the remaining space prediction model corresponding to any target time period of any type of parking lot to be predicted is used to predict the remaining space for the target time period of the parking lot to be predicted.

[0011] According to a second aspect of an embodiment of the present application, a parking lot remaining space prediction device is provided, comprising:

[0012] The clustering unit is used to cluster the parking lots to be predicted according to the clustering reference attributes of the parking lots, and obtain k types of parking lots to be predicted; wherein k ≥ 2;

[0013] The time period division unit is used to divide the time periods of various parking lots to be predicted respectively, and obtain the optimal time period division method of various parking lots to be predicted;

[0014] A training unit is used to train a remaining space prediction model for each target time period of each type of parking lot to be predicted based on the historical remaining space data of the parking lot to be predicted, and obtain the remaining space prediction model corresponding to each target time period of each type of parking lot to be predicted; wherein, for any type of parking lot to be predicted, the target time period is a time period divided according to the optimal time period division method of the parking lot to be predicted; the remaining space prediction model corresponding to any target time period of any type of parking lot to be predicted is used to perform remaining space prediction for the target time period of the parking lot to be predicted.

[0015] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor is used to execute the machine executable instructions to implement the method provided in the first aspect.

[0016] According to a fourth aspect of an embodiment of the present application, a machine-readable storage medium is provided, wherein the machine-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by a processor, the method provided in the first aspect is implemented.

[0017] According to a fifth aspect of an embodiment of the present application, a parking lot management system is provided, including: a sensor, a memory, and a processor; wherein:

[0018] The sensor is used to detect vehicles entering and exiting the parking lot, and determine the number of remaining parking spaces in the parking lot based on the detection result;

[0019] The memory is used to store the remaining space data of the parking lot and the trained remaining space prediction model; wherein the trained remaining space prediction model is trained using the method provided by the first aspect;

[0020] The processor is used to run the trained remaining space prediction model, and use the trained remaining space prediction model to predict the remaining space of the parking lot according to the historical remaining space data of the parking lot stored in the memory to obtain a remaining space prediction result.

[0021] The parking lot remaining space prediction method of the embodiment of the present application clusters the parking lots to be predicted according to the clustering reference attributes of the parking lots to obtain k types of parking lots to be predicted, and divides each type of parking lot to be predicted into time periods to obtain the optimal time period division method for each type of parking lot to be predicted. Then, based on the historical remaining space data of the parking lots to be predicted, a remaining space prediction model is trained for each target time period of each type of parking lot to be predicted, and a remaining space prediction model corresponding to each target time period of each type of parking lot to be predicted is obtained. The obtained remaining space prediction model is used to predict the remaining space for the corresponding target time period of the corresponding parking lot to be predicted. In the process of training the remaining space prediction model, not only the differences in the changing rules of the number of remaining spaces in parking lots of different categories are considered, but also the differences in the changing rules of the number of remaining spaces in the same parking lot in different time periods are considered, thereby improving the accuracy of the parking lot remaining space prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of a parking lot remaining space prediction method shown in an exemplary embodiment of the present application;

[0023] Figure 2 It is a flowchart showing a comprehensive consideration of spatiotemporal + temporal clustering according to an exemplary embodiment of the present application;

[0024] Figure 3 This is a flow chart of pattern mining difference learning between categories and attribute learning in different time periods shown in an exemplary embodiment of the present application;

[0025] Figure 4 This is a flowchart of upstream and downstream joint training prediction in different time periods shown in an exemplary embodiment of the present application;

[0026] Figure 5 It is a structural schematic diagram of a parking lot remaining space prediction device shown in an exemplary embodiment of the present application;

[0027] Figure 6 It is a schematic diagram of the hardware structure of an electronic device shown in an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the technical solutions provided by the embodiments of the present application and to make the above-mentioned purposes, features and advantages of the embodiments of the present application more obvious and understandable, the technical solutions in the embodiments of the present application are further described in detail below in conjunction with the accompanying drawings.

[0029] See also Figure 1 , is a flow chart of a parking lot remaining space prediction method provided in an embodiment of the present application, such as Figure 1 As shown, the parking lot remaining space prediction method may include the following steps:

[0030] Step S100: cluster the parking lots to be predicted according to the clustering reference attributes of the parking lots to obtain k types of parking lots to be predicted; wherein k≥2.

[0031] In the embodiment of the present application, considering that the changing patterns of the number of remaining spaces in different parking lots may be different, if the changing patterns of the number of remaining spaces in different parking lots are not considered and the same prediction model is used to predict the remaining spaces in all parking lots, the prediction effect will lead to poor.

[0032] Based on this, the clustering reference attributes of the parking lot can be determined according to the information that affects the number of remaining parking spaces in the actual scenario, and the parking lots to be predicted can be clustered according to the clustering reference attributes of the parking lot to obtain k types of parking lots to be predicted.

[0033] Exemplarily, the cluster reference attributes may include, but are not limited to, one or more of spatial location information, functional area attributes, and historical residual data.

[0034] Exemplarily, the spatial location information may include location information such as the latitude and longitude of the parking lot.

[0035] Functional area attributes may include but are not limited to residence, shopping mall or park.

[0036] Step S110: divide each type of parking lot to be predicted into time periods to obtain the optimal time period division method for each type of parking lot to be predicted.

[0037] Step S120: Based on the historical remaining space data of the parking lots to be predicted, the remaining space prediction model is trained for each target time period of each type of parking lot to be predicted, and the remaining space prediction model corresponding to each target time period of each type of parking lot to be predicted is obtained respectively; wherein, for any type of parking lot to be predicted, the target time period is a time period divided according to the optimal time period division method of the parking lot to be predicted; the remaining space prediction model corresponding to any target time period of any type of parking lot to be predicted is used to perform remaining space prediction for the target time period of the parking lot to be predicted.

[0038] In the embodiment of the present application, it is considered that there may be differences in the changing patterns of remaining spaces in the same parking lot at different time periods.

[0039] For example, for a shopping mall parking lot, there are usually fewer parking spaces available during the noon and afternoon hours, but more available spaces at other times.

[0040] For parking lots in residential areas, there are usually more vacant spaces during the day and fewer vacant spaces at night.

[0041] If the changing pattern of the number of remaining parking spaces in different time periods is not taken into account, using the same prediction model to predict the remaining parking spaces in different time periods will lead to poor prediction results.

[0042] Based on this, when the parking lots to be predicted are clustered in the above manner to obtain k types of parking lots to be predicted, each type of parking lot to be predicted can be divided into time periods to obtain the optimal time period division method for each type of parking lot to be predicted.

[0043] Among them, for any type of parking lot to be predicted, when the optimal time period division method of this type of parking lot to be predicted is determined, the optimal time period division can be performed on this type of parking lot to be predicted based on the optimal time period division method of this type of parking lot to be predicted to obtain the target time period of this type of parking lot to be predicted.

[0044] Furthermore, based on the historical remaining space data of each target time period of the parking lot to be predicted, the remaining space prediction model can be trained for each target time period of each type of parking lot to be predicted, and the remaining space prediction model corresponding to each target time period of each type of parking lot to be predicted can be obtained respectively.

[0045] The remaining space prediction model corresponding to any target time period of any type of parking lot to be predicted is used to predict the remaining space for the target time period of the parking lot to be predicted.

[0046] Exemplarily, the trained remaining space prediction model can be deployed on a remaining space prediction device (such as a remaining space prediction server). When the remaining space prediction device receives a remaining space prediction request sent by a client for any parking lot to be predicted and any time range (the time range includes at least one target time period), it can obtain the historical remaining space data of the parking lot to be predicted, and use the remaining space prediction model of the parking lot to be predicted corresponding to the target time period in the time range to perform remaining space prediction for the corresponding target time period, obtain prediction results for each target time period, and merge the prediction results for each target time period to obtain a final prediction result.

[0047] It can be seen that in Figure 1In the method flow shown, the parking lots to be predicted are clustered according to the clustering reference attributes of the parking lots to obtain k types of parking lots to be predicted, and each type of parking lot to be predicted is divided into time periods to obtain the optimal time period division method for each type of parking lot to be predicted. Then, based on the historical remaining space data of the parking lots to be predicted, the remaining space prediction model is trained for each target time period of each type of parking lot to be predicted, and the remaining space prediction model corresponding to each target time period of each type of parking lot to be predicted is obtained. The obtained remaining space prediction model is used to predict the remaining space for the corresponding target time period of the corresponding parking lot to be predicted. In the process of training the remaining space prediction model, not only the differences in the changing laws of the number of remaining spaces in parking lots of different categories are considered, but also the differences in the changing laws of the number of remaining spaces in the same parking lot in different time periods are considered, thereby improving the accuracy of parking lot remaining space prediction.

[0048] In some embodiments, the cluster reference attributes may include one or more of spatial location information, functional area attributes, and historical residual data.

[0049] The above-mentioned parking lots are clustered according to the clustering reference attributes of the parking lots, including:

[0050] In the case where the clustering reference attribute includes spatial location information, the parking lots to be predicted are clustered according to the spatial accessibility and spatial distance of the parking lots to be predicted; wherein the parking lots to be predicted that are spatially accessible and whose spatial distances are within the first distance threshold are clustered into the same category;

[0051] In the case where the clustering reference attribute includes the functional area attribute, the parking lots to be predicted are clustered according to the functional area attributes of the parking lots to be predicted; wherein the parking lots to be predicted with the same functional area attributes are clustered into the same category;

[0052] When the clustering reference attribute includes historical remaining space data, the morphological similarity of the historical remaining space data of each parking lot to be predicted is determined based on the historical remaining space data of the parking lot to be predicted, and the parking lots to be predicted are clustered based on the morphological similarity of the historical remaining space data of each parking lot to be predicted.

[0053] For example, considering that the spatial location and functional area attributes of a parking lot may affect the variation pattern of the number of remaining parking spaces in the parking lot, in addition, the morphological similarity between the variation curves of the remaining space data of different parking lots can also represent whether the variation patterns of the remaining space data of the corresponding parking lots are similar.

[0054] Therefore, parking lots may be clustered based on one or more of spatial location information, functional area attributes, and historical remaining parking space data.

[0055] In one example, when the clustering reference attribute includes spatial location information, the spatial accessibility and spatial distance between different parking lots to be predicted can be determined based on the spatial location information of the parking lots to be predicted, and the parking lots to be predicted can be clustered based on the spatial accessibility and spatial distance between the different parking lots to be predicted.

[0056] Exemplarily, parking lots to be predicted that are spatially accessible and whose spatial distance is within a preset distance threshold (which may be referred to as a first distance threshold) may be grouped into the same category.

[0057] For example, the spatial distance between parking lots may be determined based on static map lines.

[0058] In one example, in the case of clustering reference attributes and functional area attributes, the functional area attributes of each parking lot to be predicted may be determined, and the parking lots to be predicted may be clustered according to the functional area attributes of each parking lot to be predicted.

[0059] Exemplarily, parking lots to be predicted with the same functional area attributes may be grouped into the same category.

[0060] In one example, when the clustering reference attribute includes historical remaining space data, the morphological similarity of the historical remaining space data of each parking lot to be predicted may be determined based on the historical remaining space data of the parking lot to be predicted.

[0061] Exemplarily, for any parking lot to be predicted, a remaining space quantity-time relationship curve can be determined based on the historical remaining space data of the parking lot to be predicted, and further, the morphological similarity of the historical remaining space data of the parking lot to be predicted can be determined based on the remaining space quantity-time relationship curves of different parking lots to be predicted.

[0062] For example, for two parking lots to be predicted, the Euclidean distance can be used to determine the morphological similarity of the historical remaining space data of the two parking lots to be predicted by calculating the sum of squares of the difference in the number of remaining spaces at each time point in the remaining space number-time relationship curve of the two parking lots to be predicted.

[0063] For another example, for two parking lots to be predicted, the shape similarity of the historical remaining space data of the parking lots to be predicted can be determined based on whether the shapes of the remaining space quantity-time relationship curves of the two parking lots to be predicted are similar. For example, the number of remaining spaces in a parking lot may present a fluctuating pattern, and the shape similarity analysis can detect whether the fluctuation patterns in different time periods are consistent.

[0064] For example, according to the morphological similarity of the historical remaining space data of each parking lot to be predicted, the parking lots to be predicted that meet the morphological similarity requirement (for example, the Euclidean distance is less than a preset Euclidean distance threshold) can be grouped into one category.

[0065] In one example, the cluster reference attributes include spatial location information, functional area attributes, and historical residual data;

[0066] Based on the clustering reference attributes of the parking lots, the parking lots to be predicted are clustered to obtain k types of parking lots to be predicted, including:

[0067] According to the spatial accessibility and spatial distance of the parking lot to be predicted, the parking lot to be predicted is clustered to obtain n types of parking lots to be predicted; n ≥ 2;

[0068] According to the functional area attributes of each parking lot to be predicted, the n types of parking lots to be predicted are clustered again to obtain m types of parking lots to be predicted; wherein m≤n;

[0069] According to the historical remaining space data of each parking lot to be predicted, the morphological similarity of the historical remaining space data of each parking lot to be predicted in each type of parking lot to be predicted in the m types of parking lots to be predicted is determined respectively, and according to the morphological similarity of the historical remaining space data of each parking lot to be predicted in each type of parking lot to be predicted, each parking lot to be predicted in each type of parking lot to be predicted is clustered to obtain k types of parking lots to be predicted; k≥m.

[0070] For example, in order to improve the accuracy and rationality of clustering, parking lots may be clustered based on their spatial location information, functional area attributes, and historical remaining space data.

[0071] For example, in the process of clustering parking lots, the parking lots to be predicted may be clustered according to their spatial accessibility and spatial distance to obtain n types of parking lots to be predicted.

[0072] For the obtained n types of parking lots to be predicted, the functional area attributes of the n types of parking lots to be predicted can be added according to the functional area attributes of each parking lot to be predicted, and clustering is performed again to obtain m types of parking lots to be predicted.

[0073] For any type of the m types of parking lots to be predicted, the morphological similarity of the historical remaining space data of each parking lot to be predicted in this type can be determined based on the historical remaining space data of each parking lot to be predicted in this type, and based on the morphological similarity of the historical remaining space data of each parking lot to be predicted in this type, each parking lot to be predicted in this type can be clustered again.

[0074] For the m types of parking lots to be predicted, intra-class clustering can be performed again in the above manner to finally obtain k types of parking lots to be predicted.

[0075] In some embodiments, the above-mentioned division of time periods for each type of parking lot to be predicted to obtain the optimal time period division method for each type of parking lot to be predicted may include:

[0076] Based on the functional area data and the historical remaining space data of various parking lots to be predicted, pattern mining and learning are performed on various parking lots to be predicted to determine the inter-class differences between the various parking lots to be predicted;

[0077] According to the inter-class differences between various parking lots to be predicted and the historical remaining space data of various parking lots to be predicted, various parking lots to be predicted are divided into time periods, and the optimal time period division method for various parking lots to be predicted is obtained.

[0078] Exemplarily, when the clustering of the parking lots to be predicted is achieved as described above, the differences between different classes can be further learned, the difference attributes of the categories can be highly distinguished, and adaptive time period selection can be further performed for different time periods within the class. By combining the differences between classes and the differences between different time periods within the class, the data can be deeply differentiated so that the model can better learn the rules of different classes and different time periods, thereby better predicting future data and improving prediction accuracy.

[0079] Accordingly, when clustering of parking lots to be predicted is achieved as described above, pattern mining and learning can be performed on each type of parking lot to be predicted based on the functional area data and the historical remaining space data of each type of parking lot to be predicted to determine the inter-class differences between the various types of parking lots to be predicted.

[0080] Furthermore, according to the inter-class differences between the various parking lots to be predicted and the historical remaining space data of the various parking lots to be predicted, the various parking lots to be predicted can be divided into time periods to obtain the optimal time period division method for the various parking lots to be predicted.

[0081] In one example, the inter-class differences between the various parking lots to be predicted are determined based on the functional area attributes of the various parking lots to be predicted;

[0082] The above method of dividing each type of parking lot to be predicted into time periods based on the inter-class differences between each type of parking lot to be predicted and the historical remaining space data of each type of parking lot to be predicted, and obtaining the optimal time period division method for each type of parking lot to be predicted, may include:

[0083] According to the functional area attributes of various parking lots to be predicted, the morphological fluctuations of the historical remaining space data of various parking lots to be predicted, and the initial time period division method of various parking lots to be predicted, the regression model is used to determine the optimal time period division method of various parking lots to be predicted.

[0084] Exemplarily, the functional area attributes of each type of parking lot to be predicted may be used to characterize the inter-class differences between the various types of parking lots to be predicted.

[0085] For example, a regression model may be used to implement an adaptive time period selection method and determine an optimal time period division method based on functional area attributes and fluctuations in the form of historical parking lot remaining space data.

[0086] Exemplarily, for any type of parking lot to be predicted, the optimal time period division method for the type of parking lot to be predicted can be determined using a regression model based on the functional area attributes of the type of parking lot to be predicted, the morphological fluctuations of historical remaining space data, and the initial time period division method.

[0087] In one example, the above-mentioned optimal time period division method for each type of parking lot to be predicted is determined by using a regression model based on the functional area attributes of each type of parking lot to be predicted, the morphological fluctuation of the historical remaining space data of each type of parking lot to be predicted, and the initial time period division method of each type of parking lot to be predicted, which may include:

[0088] Input the functional area attributes of each type of parking lot to be predicted and the initial time period division method of each type of parking lot to be predicted into the tree model to obtain the recommended time period division method of each type of parking lot to be predicted;

[0089] Using the regression model, the effectiveness of the current recommended time period division method for each type of parking lot to be predicted is evaluated;

[0090] When the effect evaluation result does not meet the requirements, the functional area attributes of each type of parking lot to be predicted and the current recommended time period division method of each type of parking lot to be predicted are input into the tree model to obtain a new recommended time period division method of each type of parking lot to be predicted;

[0091] When the effect evaluation results meet the requirements, the currently recommended time period division method for each type of parking lot to be predicted is determined as the optimal time period division method for each type of parking lot to be predicted.

[0092] For example, a tree model may be used to determine a recommended time period division method for a parking lot, and a regression model may be used to evaluate the effectiveness of the time period division method for the parking lot.

[0093] Accordingly, the functional area attributes of each type of parking lot to be predicted and the initial time period division method of each type of parking lot to be predicted can be input into the tree model to obtain the recommended time period division method of each type of parking lot to be predicted.

[0094] For example, when the recommended time period division methods for various types of parking lots to be predicted are obtained, a regression model can be used to evaluate the effectiveness of the current recommended time period division methods for various types of parking lots to be predicted, and determine whether the effectiveness evaluation results meet the requirements.

[0095] When the effect evaluation results do not meet the requirements, the recommended time period division method can be iteratively updated, that is, the functional area attributes of each type of parking lot to be predicted and the current recommended time period division method of each type of parking lot to be predicted are input into the tree model to obtain a new recommended time period division method for each type of parking lot to be predicted.

[0096] When the effect evaluation results meet the requirements, the currently recommended time period division method for each type of parking lot to be predicted is determined as the optimal time period division method for each type of parking lot to be predicted.

[0097] In some embodiments, the above-mentioned training of the remaining space prediction model for each target time period of each type of parking lot to be predicted based on the historical remaining space data of each target time period of the parking lot to be predicted may include:

[0098] For any target time period of any type of parking lot to be predicted, a remaining space prediction model corresponding to the target time period of the type of parking lot to be predicted is trained based on the training data of the target time period of the type of parking lot to be predicted;

[0099] The training data of the target time period of the parking lot to be predicted includes: historical remaining space data of the parking lot to be predicted, and reference predicted remaining space data of adjacent time periods.

[0100] Exemplarily, in order to optimize the training effect of the remaining space prediction model, for any type of parking lot to be predicted, during the training process of the remaining space prediction model of that type of parking lot to be predicted, the training data may include not only the historical remaining space data of that type of parking lot to be predicted, but also the reference remaining space prediction data of adjacent time periods.

[0101] Exemplarily, the reference remaining space prediction data for adjacent time periods may include remaining space prediction results for similar parking lots in adjacent time periods and / or remaining space prediction results for different parking lots in adjacent time periods.

[0102] In one example, the reference residual position prediction data of adjacent time periods includes one or more of the following:

[0103] The remaining space prediction results of the upstream time period of the parking lot to be predicted of this type, and the remaining space prediction results of the upstream time period of other types of parking lots to be predicted that are most adjacent to the parking lot to be predicted of this type;

[0104] The most adjacent other types of parking lots to be predicted refers to other types of parking lots to be predicted whose spatial distance to the parking lot to be predicted is less than the second distance threshold and whose spatial distance to the parking lot to be predicted is the shortest.

[0105] Exemplarily, different models can be used to train different time periods, and the upstream and downstream of different time periods can be jointly influenced. The residual prediction results of the spatially similar (belonging to the same category after clustering) upstream time period (such as the adjacent previous time period) can be transferred to the downstream time period to assist the residual prediction of the downstream time period, so as to improve the prediction accuracy of the downstream time period.

[0106] In addition, the residual position prediction result of the upstream time period can also be passed to the inter-class downstream time period based on the inter-class spatial position and attribute judgment, so as to improve the residual position prediction accuracy of the inter-class downstream time period.

[0107] Accordingly, for any target time period of any type of parking lot to be predicted, the training data of the target time period of the parking lot to be predicted may include: historical remaining space data of the parking lot to be predicted, and one or more of the following:

[0108] The remaining space prediction results of the upstream time period of the parking lot to be predicted of this type, and the remaining space prediction results of the upstream time period of other types of parking lots to be predicted that are most adjacent to the parking lot to be predicted of this type.

[0109] Exemplarily, the most adjacent other type of parking lot to be predicted refers to the other type of parking lot to be predicted having the shortest spatial distance to the parking lot to be predicted among the other types of parking lots to be predicted whose spatial distance to the parking lot to be predicted is less than the second distance threshold.

[0110] In one example, for two types of parking lots to be predicted, an average value of spatial distances between different types of parking lots in the two types of parking lots to be predicted may be determined as the distance between the two types of parking lots to be predicted.

[0111] For example, assuming that Class A parking lot includes parking lots A1 and A1, and Class B parking lot includes parking lots B1, B2 and B3, the spatial distance between A1 and B1, B2 and B3, as well as the spatial distance between A2 and B1, B2 and B3 can be determined respectively, and the average of the six spatial distances can be determined as the spatial distance between Class A parking lot and Class B parking lot.

[0112] In another example, for two types of parking lots to be predicted, the maximum distance (or minimum distance) among the spatial distances between parking lots of different types in the two types of parking lots to be predicted may be determined as the distance between the two types of parking lots to be predicted.

[0113] Still taking the above example, the maximum value (or minimum value) of the six spatial distances may be determined as the distance between the two types of parking lots to be predicted.

[0114] It should be noted that, in the embodiment of the present application, for any parking lot, when determining the spatial distance between other types of parking lots and this type of parking lot, the spatial distance can also be corrected based on the predicted remaining space data of other types of parking lots, for example, by multiplying it by a correction coefficient, which can be negatively correlated with the number of remaining spaces.

[0115] For example, assuming that for parking lot A1, the calculated actual spatial distances between parking lot B1 and parking lot B2 and parking lot A1 are the same, but the predicted number of remaining spaces for parking lot B1 is greater than the predicted number of remaining spaces for parking lot B2, then the corrected spatial distance between parking lot B1 and parking lot A1 may be smaller than the corrected spatial distance between parking lot B2 and parking lot A1.

[0116] Accordingly, when performing relevant processing based on the spatial distance between parking lots in the above embodiment, the above corrected spatial distance may be used.

[0117] Exemplarily, for any target time period of any type of parking lot to be predicted, the remaining space prediction model corresponding to the target time period of the type of parking lot to be predicted is trained based on the training data of the target time period of the type of parking lot to be predicted.

[0118] For example, during the training of the remaining space prediction model, a training set and a test set may be constructed based on historical remaining space data of the parking lot.

[0119] In the process of training the residual prediction model using the training set, the model training loss can be determined based on the difference between the predicted results of the residual prediction model and the actual results included in the historical residual data, and the model can be feedback-tuned based on the determined training loss.

[0120] After the training is completed, the prediction effect of the residual prediction model obtained through training is tested using the test set; if the test result does not meet the conditions, training is performed again until the test result meets the conditions.

[0121] For example, assuming that the training data in the acquired historical remaining space data includes the historical remaining space data of the parking lot from October 1, 24 to October 11, 24, then the historical remaining space data of the parking lot from October 1, 24 to October 10, 24 can be used as training samples, and the historical remaining space data of the parking lot in each time period on October 11, 24 can be used as sample labels to predict the remaining space prediction model.

[0122] For example, assuming a time period of 12 hours, the training samples can be used to train the remaining space prediction model to obtain the remaining space prediction results of the remaining space prediction model for 0-12 o'clock on October 11, 24, and the remaining space prediction results for 12-24 o'clock (different time periods can correspond to different remaining space prediction models), and compare the actual remaining space data of the parking lot from 0-12 o'clock on October 11, 24, and the actual remaining space data from 12-24 o'clock, determine the training loss, and perform feedback and optimization on the model based on the training loss.

[0123] It should be noted that, in the process of training the remaining position prediction model, the remaining position prediction model can also be trained to predict the number of remaining positions for multiple days.

[0124] For example, the historical remaining space data of the parking lot from October 1, 24 to October 10, 24 can be used as training samples, and the historical remaining space data of the parking lot in each time period from October 11, 24 to October 17, 24 can be used as sample labels. The remaining space prediction model is used to predict the remaining space data in each time period from October 11, 24 to October 17, 24. Then, after the model training is completed, the trained remaining space prediction model can be used to predict the remaining space data for the next 7 consecutive days.

[0125] For the test implementation process of the residual prediction model, please refer to the training process.

[0126] In an embodiment of the present application, when the remaining space prediction models corresponding to each target time period of each type of parking lot to be predicted are obtained in the manner described in the above embodiments, the trained remaining space prediction models can be used to predict the remaining space for the corresponding target time period of the corresponding parking lot to be predicted.

[0127] Exemplarily, in the process of performing residual prediction (actual reasoning) using the residual prediction model, the form of input data of the residual prediction model is consistent with the form of input data during the residual prediction model training process.

[0128] Still taking the above model training example as an example, after obtaining the trained remaining space prediction model in the above manner, assuming that it is necessary to predict the remaining space data of the parking lot from 0 to 12 o'clock on December 11, 24, the historical remaining space data from December 1, 24 to December 10, 24 can be obtained. Based on the obtained historical remaining space data, the trained remaining space prediction model is used to predict the remaining space data of the parking lot from 0 to 12 o'clock on December 11, 24.

[0129] In some embodiments, the parking lot remaining space prediction scheme provided in the embodiments of the present application may also include:

[0130] Obtain historical remaining space data of the parking lot to be predicted, and aggregate and count the historical remaining space data of each parking lot to be predicted according to a preset statistical period;

[0131] For any parking lot to be predicted, if there is missing data within the specified working time range, and the ratio of the duration of missing data to the preset statistical period does not exceed the preset threshold, the missing data will be supplemented based on the remaining data of the same period in history;

[0132] For any parking lot to be predicted, when there are data anomalies in the aggregated statistical results of the parking lot to be predicted, the abnormal data are corrected based on the remaining data of the adjacent statistical period and the remaining data of the same historical period.

[0133] Exemplarily, in order to improve the training efficiency of the remaining space prediction model and improve the prediction accuracy of the remaining space prediction model, the historical remaining space data of the parking lot to be predicted may be preprocessed.

[0134] Exemplarily, the above preprocessing may include clustering of residual data to reduce the amount that needs to be processed; supplementation of missing data and correction of abnormal data to improve data quality, thereby improving the prediction accuracy of the residual prediction model.

[0135] For example, for the acquired historical remaining space data of the parking lot to be predicted, the historical remaining space data of each parking lot to be predicted may be aggregated and counted according to a preset statistical period (the specific value may be set according to the actual scenario).

[0136] For example, assuming that the preset statistical cycle is 15 minutes, the remaining space data of each parking lot to be predicted can be aggregated and counted to obtain a data record every 15 minutes. Among them, each parking lot has 4 records per hour and 96 data records per day.

[0137] Exemplarily, for any parking lot to be predicted, if there is missing data within the specified working time range, and the ratio of the duration of the data missing to the preset statistical period does not exceed the preset threshold, the missing data can be supplemented based on the remaining data of the same historical period.

[0138] Exemplarily, the above-mentioned designated working time range may be a time period when vehicles enter and exit relatively frequently, and may be set manually based on experience.

[0139] For example, the specified working hours range can be from 5 am to 12 pm.

[0140] For example, taking a data record of 15 minutes as an example, assuming that the data records from 8:00 to 8:15 in the specified working range in the historical remaining parking lot data are missing, the data of the same historical period (such as 8:00 to 8:15 of the previous day, and 8:00 to 8:15 of the previous day) can be interpolated and filled to ensure that there is no gap in the overall historical data.

[0141] It should be noted that if there is missing data within the non-specified working hours and the ratio of the duration of the data missing to the preset statistical period does not exceed the preset threshold, data can be supplemented in the above manner, or data from the previous statistical period can be used for filling.

[0142] In addition, in the case where there is data missing and the ratio of the duration of the data missing to the preset statistical period exceeds a preset threshold, other processing measures may be taken, such as manual intervention.

[0143] Exemplarily, for any parking lot to be predicted, when there are data anomalies in the aggregated statistical results of the parking lot to be predicted, the abnormal data is corrected based on the remaining data of the adjacent statistical period and the remaining data of the same historical period.

[0144] Exemplarily, for a statistical period, when the statistical result of the statistical period changes by more than a preset ratio threshold relative to the statistical result of the previous statistical period, it can be determined that there is a data anomaly.

[0145] Taking into account the actual scenario, due to certain special reasons, such as holding large-scale events, etc., the residual data in certain statistical periods may change greatly. Therefore, when it is determined that the statistical results of a statistical period have changed by more than a preset proportion threshold compared with the statistical results of the previous statistical period, it can be further determined whether there is a specific activity arrangement (which can be configured manually). If there is no specific activity arrangement, it is determined that there is a data anomaly.

[0146] Exemplarily, when it is determined that data anomalies exist, the abnormal data may be corrected based on the residual data of adjacent statistical periods of the statistical period corresponding to the abnormal data and the residual data of the same historical period.

[0147] For example, assuming that the remaining data of a parking lot from 8:00 to 8:15 on a certain day is abnormal, the abnormal data can be corrected based on the remaining data of the previous statistical cycles and the remaining data from 8:00 to 8:15 on the previous day or days.

[0148] In order to enable those skilled in the art to better understand the technical solutions provided in the embodiments of the present application, the technical solutions provided in the embodiments of the present application are described below with reference to specific examples.

[0149] In this embodiment, a parking lot remaining space prediction solution combining spatiotemporal clustering and multi-expert hybrid is provided, including processing of historical parking lot remaining space data, spatiotemporal clustering of parking lots and time series clustering of remaining space data, pattern mining difference learning between parking lot categories and attribute learning in different time periods, and joint training prediction of upstream and downstream in different time periods. After obtaining the historical parking lot remaining space data, the above four processes are respectively carried out to obtain the parking lot remaining space prediction data for one or more days in the future.

[0150] The above four processes are described below respectively.

[0151] 1. Processing of historical parking lot remaining space data.

[0152] For the historical parking lot remaining space data, the following methods can be used for preprocessing:

[0153] 1.1. Historical data aggregation statistics: When the historical remaining data is obtained, the remaining data is aggregated and counted according to different parking lots, and a data record is obtained every 15 minutes. Each parking lot has 4 records per hour, and there are 96 data records per day.

[0154] 1.2. Data missing preprocessing: For data missing within a short period of time (the ratio to the statistical period does not exceed the threshold) and within the normal working time period (which can be manually configured), the historical data of the same period can be used for interpolation and filling to ensure that there is no gap in the overall historical data.

[0155] 1.3. Data anomaly preprocessing: For historical residual data that suddenly drops or rises in a short period of time, combined with actual business scenarios, and after confirming that no other large-scale events or other events occur during this period, perform anomaly processing on this part of the data. For example, you can use the data from the previous period and the historical period to combine the data for anomaly processing.

[0156] 2. Spatiotemporal clustering of parking lots and temporal clustering of residual data.

[0157] Since the parking lot spatial location and functional area attributes will affect the change pattern of parking lot remaining space, the parking lot spatial location information (such as longitude and latitude) can be combined with the geographical attributes of the parking lot area to cluster the parking lot space. According to the distribution of parking in the spatial location, the basic spatial accessibility and spatial accessibility distance clustering are first performed to cluster the parking lots into n categories; then, combined with the geographical attributes of the parking lot area (or functional area attributes, such as residential, shopping malls, parks, etc.), the functional attributes of the n categories are added, and the n types of parking lots are clustered again to obtain m types of parking lots. Furthermore, the historical remaining space data of parking lots in different categories are obtained, and the parking space remaining space data in each category are clustered in time series by using morphological similarity. The parking lots in each category are clustered into different categories according to morphological similarity, and finally, k types of parking lots are obtained based on spatiotemporal clustering and time series clustering.

[0158] For example, Figure 2 As shown in the figure, the optimization steps of parking lot spatiotemporal clustering and residual data temporal clustering can be as follows:

[0159] 1) Data processing of parking lot space location and spatial functional area attributes (POI);

[0160] 2) Calculate the spatial accessibility and reachable distance of parking lot locations, and cluster them according to spatial accessibility and reachable distance;

[0161] 3) Recalculate the regional categories by combining spatial reachability distance and functional area attributes;

[0162] 4) For any type of parking lot, perform morphological similarity calculation based on historical remaining space data, use morphological similarity to perform time series clustering on the remaining parking space data in each category, and cluster the parking lots in each category into different categories based on morphological similarity;

[0163] 5) Get the final clustering result (such as the k categories mentioned above).

[0164] 3. Pattern mining difference learning between parking lot categories and attribute learning in different time periods.

[0165] Since the patterns of changes in remaining parking spaces in different time periods in the same parking lot may also be different, therefore, when the parking lots are clustered from a spatiotemporal perspective in the above manner, the differences between different classes can be further learned, the difference attributes of the categories can be highly distinguished, and adaptive time period selection can be further performed for different time periods within a class. By combining the differences between classes and the differences between different time periods within a class, the data can be deeply distinguished, allowing the model to better learn the patterns of different classes and different time periods, thereby better predicting future data.

[0166] Exemplarily, pattern mining difference learning between parking lot categories and attribute learning in different time periods may include: pattern mining difference learning between categories and attribute learning in different time periods; wherein:

[0167] 3.1. Pattern mining and difference learning between categories: After the above-mentioned spatiotemporal clustering combined with temporal clustering, multiple categories of parking lots are finally obtained. Then, pattern mining and learning are performed between each category in combination with the area (functional attribute) to which each category belongs, and the differences between different categories are learned, so as to obtain the difference attributes that are different from different categories. For example, functional area [entertainment area, commercial area, residential area, park leisure area, etc.].

[0168] 3.2. Attribute learning in different time periods: Attribute learning in different time periods is performed between classes to obtain high-weight attributes in different time periods between different classes.

[0169] For example, according to the functional area attributes and the fluctuations in the form of historical parking space data, a regression model can be used to adaptively select time periods, with parking space data and initialization time periods as input, to ultimately obtain the time period division in each category and the high-weight features of different time periods in each category. At this point, the characteristic laws exhibited by different time periods between classes and within classes are obtained, and the optimal time period division method for each category is determined.

[0170] like Figure 3 As shown in the figure, the steps of learning the difference between categories and learning the attributes of different time periods can be as follows:

[0171] 1) Build a parallelized category input network;

[0172] 2) Input multi-category sequence data and functional area data into pattern mining to obtain the attribute dimension characteristics of each category’s functional area;

[0173] 3) Input the categories of the functional area attributes and the initialization time periods into the tree model to divide them into different time periods;

[0174] 4) Determine the time period differences and effects through regression model and iterate the third step;

[0175] 5) Finally, the optimal time period division method within the class is obtained.

[0176] 4. Upstream and downstream joint training prediction in different time periods.

[0177] Different models can be used to train different time periods. At the same time, the model conducts joint influence transmission for upstream and downstream of different time periods, and transmits the residual prediction results of the upstream time period (spatial similarity) to the downstream time period (spatial similarity) to assist the residual prediction of the downstream time period, so as to improve the prediction accuracy of the downstream time period.

[0178] In addition, the residual position prediction result of the upstream time period can also be passed to the inter-class downstream time period based on the inter-class spatial position and attribute judgment, so as to improve the residual position prediction accuracy of the inter-class downstream time period.

[0179] For example, Figure 4 As shown, the upstream and downstream joint training prediction optimization steps in different time periods can be as follows:

[0180] 1) Construct data of different time periods in different categories (divided according to the optimal time period division method, that is, the above-mentioned target time period), and connect each time period sequentially;

[0181] 2) Input data from different time periods into multiple models and select the best model for each time period;

[0182] 3) Add the prediction results of each time period to the data set of the next time period to transfer the results of the upstream time period;

[0183] 4) Obtain the optimal model results for each time period.

[0184] When the optimal model for each time period is obtained, the model performs predictive reasoning according to each time period during reasoning, and finally obtains the results of all time periods. The results of all time periods are combined to obtain the final prediction result.

[0185] The method provided by the present application is described above. The device provided by the present application is described below:

[0186] See also Figure 5 , is a schematic diagram of the structure of a parking lot remaining space prediction device provided in an embodiment of the present application, such as Figure 5 As shown, the parking lot remaining space prediction device may include:

[0187] A clustering unit 510 is used to cluster the parking lots to be predicted according to the clustering reference attributes of the parking lots to obtain k types of parking lots to be predicted; wherein k≥2;

[0188] The time period division unit 520 is used to divide the time periods of each type of parking lot to be predicted, and obtain the optimal time period division method for each type of parking lot to be predicted;

[0189] The training unit 530 is used to train the remaining space prediction model for each target time period of each type of parking lot to be predicted based on the historical remaining space data of the parking lot to be predicted, and obtain the remaining space prediction model corresponding to each target time period of each type of parking lot to be predicted; wherein, for any type of parking lot to be predicted, the target time period is the time period obtained by dividing according to the optimal time period division method of the parking lot to be predicted; the remaining space prediction model corresponding to any target time period of any type of parking lot to be predicted is used to perform remaining space prediction for the target time period of the parking lot to be predicted.

[0190] In some embodiments, the cluster reference attributes include one or more of spatial location information, functional area attributes, and historical residual data;

[0191] The clustering unit 510 clusters the parking lots to be predicted according to the clustering reference attributes of the parking lots, including:

[0192] In the case where the cluster reference attribute includes spatial location information, clustering the parking lots to be predicted according to spatial accessibility and spatial distance of the parking lots to be predicted; wherein the parking lots to be predicted that are spatially accessible and whose spatial distances are within a first distance threshold are clustered into the same category;

[0193] In the case where the cluster reference attribute includes a functional area attribute, clustering the parking lots to be predicted according to the functional area attributes of the parking lots to be predicted; wherein the parking lots to be predicted with the same functional area attributes are clustered into the same category;

[0194] In the case where the clustering reference attribute includes historical remaining space data, the morphological similarity of the historical remaining space data of each parking lot to be predicted is determined based on the historical remaining space data of the parking lot to be predicted, and the parking lots to be predicted are clustered based on the morphological similarity of the historical remaining space data of each parking lot to be predicted.

[0195] In some embodiments, the cluster reference attributes include spatial location information, functional area attributes, and historical residual data;

[0196] The clustering unit 510 clusters the parking lots to be predicted according to the clustering reference attributes of the parking lots to obtain k types of parking lots to be predicted, including:

[0197] Clustering the parking lots to be predicted according to their spatial accessibility and spatial distance to obtain n types of parking lots to be predicted; n≥2;

[0198] According to the functional area attributes of each parking lot to be predicted, the n types of parking lots to be predicted are clustered again to obtain m types of parking lots to be predicted; wherein m≤n;

[0199] According to the historical remaining space data of each parking lot to be predicted, the morphological similarity of the historical remaining space data of each parking lot to be predicted in each type of parking lot to be predicted in the m types of parking lots to be predicted is determined respectively, and according to the morphological similarity of the historical remaining space data of each parking lot to be predicted in each type of parking lot to be predicted, each parking lot to be predicted in each type of parking lot to be predicted is clustered to obtain k types of parking lots to be predicted; k≥m.

[0200] In some embodiments, the method of dividing each type of parking lot to be predicted into time periods to obtain the optimal time period division method for each type of parking lot to be predicted includes:

[0201] Based on the functional area data and the historical remaining space data of various parking lots to be predicted, pattern mining and learning are performed on various parking lots to be predicted to determine the inter-class differences between the various parking lots to be predicted;

[0202] According to the inter-class differences between various parking lots to be predicted and the historical remaining space data of various parking lots to be predicted, various parking lots to be predicted are divided into time periods, and the optimal time period division method for various parking lots to be predicted is obtained.

[0203] In some embodiments, the inter-class differences between the various types of parking lots to be predicted are determined based on the functional area attributes of the various types of parking lots to be predicted;

[0204] The time period division unit 520 divides the various parking lots to be predicted into time periods according to the inter-class differences between the various parking lots to be predicted and the historical remaining space data of the various parking lots to be predicted, and obtains the optimal time period division method for the various parking lots to be predicted, including:

[0205] According to the functional area attributes of various parking lots to be predicted, the morphological fluctuations of the historical remaining space data of various parking lots to be predicted, and the initial time period division method of various parking lots to be predicted, the regression model is used to determine the optimal time period division method of various parking lots to be predicted.

[0206] In some embodiments, the time period division unit 520 determines the optimal time period division method for each type of parking lot to be predicted by using a regression model according to the functional area attributes of each type of parking lot to be predicted, the morphological fluctuation of the historical remaining space data of each type of parking lot to be predicted, and the initial time period division method of each type of parking lot to be predicted, including:

[0207] Input the functional area attributes of each type of parking lot to be predicted and the initial time period division method of each type of parking lot to be predicted into the tree model to obtain the recommended time period division method of each type of parking lot to be predicted;

[0208] Using the regression model, the effectiveness of the current recommended time period division method for each type of parking lot to be predicted is evaluated;

[0209] When the effect evaluation result does not meet the requirements, the functional area attributes of each type of parking lot to be predicted and the current recommended time period division method of each type of parking lot to be predicted are input into the tree model to obtain a new recommended time period division method of each type of parking lot to be predicted;

[0210] When the effect evaluation results meet the requirements, the currently recommended time period division method for each type of parking lot to be predicted is determined as the optimal time period division method for each type of parking lot to be predicted.

[0211] In some embodiments, the training unit 530 performs a remaining space prediction model training for each target time period of each type of parking lot to be predicted based on the historical remaining space data of each target time period of the parking lot to be predicted, including:

[0212] For any target time period of any type of parking lot to be predicted, a remaining space prediction model corresponding to the target time period of the type of parking lot to be predicted is trained based on the training data of the target time period of the type of parking lot to be predicted;

[0213] The training data of the target time period of the parking lot to be predicted includes: historical remaining space data of the parking lot to be predicted, and reference remaining space prediction data of adjacent time periods.

[0214] In some embodiments, the reference residual position prediction data of the adjacent time periods includes one or more of the following:

[0215] The remaining space prediction results of the upstream time period of the parking lot to be predicted of this type, and the remaining space prediction results of the upstream time period of other types of parking lots to be predicted that are most adjacent to the parking lot to be predicted of this type;

[0216] The most adjacent other types of parking lots to be predicted refers to other types of parking lots to be predicted whose spatial distance to the parking lot to be predicted is less than the second distance threshold and whose spatial distance to the parking lot to be predicted is the shortest.

[0217] In some embodiments, the parking lot remaining space prediction device may further include:

[0218] A data preprocessing unit is used to obtain historical remaining space data of the parking lot to be predicted, and to aggregate and count the historical remaining space data of each parking lot to be predicted according to a preset statistical period;

[0219] For any parking lot to be predicted, if there is data missing within the specified working time range, and the ratio of the duration of data missing to the preset statistical period does not exceed the preset threshold, the missing data is supplemented based on the remaining data of the same period in history;

[0220] For any parking lot to be predicted, when there are data anomalies in the aggregated statistical results of the parking lot to be predicted, the abnormal data are corrected based on the remaining data of the adjacent statistical period and the remaining data of the same historical period.

[0221] An embodiment of the present application provides an electronic device, including a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor is used to execute the machine executable instructions to implement the parking lot remaining space prediction method described above.

[0222] See also Figure 6 , is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. The electronic device may include a processor 601 and a memory 602 storing machine executable instructions. The processor 601 and the memory 602 may communicate via a system bus 603. Furthermore, by reading and executing the machine executable instructions corresponding to the parking lot remaining space prediction logic in the memory 602, the processor 601 may execute the parking lot remaining space prediction method described above.

[0223] The memory 602 mentioned herein may be any electronic, magnetic, optical or other physical storage device that may contain or store information, such as executable instructions, data, etc. For example, the machine-readable storage medium may be: RAM (RadomAccess Memory, random access memory), volatile memory, non-volatile memory, flash memory, storage drive (such as hard disk drive), solid state drive, any type of storage disk (such as optical disk, DVD, etc.), or similar storage medium, or a combination thereof.

[0224] In some embodiments, a machine-readable storage medium is also provided. Figure 6 The memory 602 in the machine readable storage medium stores machine executable instructions, and when the machine executable instructions are executed by the processor, the parking lot remaining space prediction method described above is implemented. For example, the storage medium can be ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

[0225] The embodiment of the present application also provides a parking lot management system including: a sensor, a memory and a processor; wherein:

[0226] The sensor is used to detect vehicles entering and exiting the parking lot, and determine the number of remaining parking spaces in the parking lot based on the detection result;

[0227] The memory is used to store the remaining space data of the parking lot and the trained remaining space prediction model; wherein the trained remaining space prediction model is trained using the parking lot remaining space prediction method described above;

[0228] The processor is used to run the trained remaining space prediction model, and use the trained remaining space prediction model to predict the remaining space of the parking lot according to the historical remaining space data of the parking lot stored in the memory to obtain a remaining space prediction result.

[0229] Exemplarily, the above-mentioned sensors may include but are not limited to one or more of cameras, radars and infrared sensors.

[0230] Exemplarily, the trained remaining space prediction model includes a remaining space prediction model corresponding to each target time period of the parking lot.

[0231] When performing the remaining space prediction using the trained remaining space prediction model, the processor may respectively use the remaining space prediction models corresponding to different target time periods to determine the remaining space prediction result of the parking lot in the corresponding target time period.

[0232] Exemplarily, the processor may combine the remaining space prediction results of each target time period of the parking lot to obtain a final remaining space prediction result of the parking lot (including remaining space prediction results of different target time periods).

Claims

1. A parking lot remaining space prediction method, characterized in that: include: According to the clustering reference attributes of the parking lot, the parking lot to be predicted is clustered to obtain k types of parking lots to be predicted; where k ≥ 2; Divide each type of parking lot to be predicted into time periods, and obtain the optimal time period division method for each type of parking lot to be predicted; According to the historical remaining space data of the parking lot to be predicted, the remaining space prediction model is trained for each target time period of each type of parking lot to be predicted, and the remaining space prediction model corresponding to each target time period of each type of parking lot to be predicted is obtained respectively; wherein, for any type of parking lot to be predicted, the target time period is a time period obtained by dividing according to the optimal time period division method of the parking lot to be predicted; the remaining space prediction model corresponding to any target time period of any type of parking lot to be predicted is used to perform remaining space prediction for the target time period of the parking lot to be predicted; The method of dividing each type of parking lot to be predicted into time periods to obtain the optimal time period division method for each type of parking lot to be predicted includes: Based on the functional area data and the historical remaining space data of various parking lots to be predicted, pattern mining and learning are performed on various parking lots to be predicted to determine the inter-class differences between the various parking lots to be predicted; According to the inter-class differences between various parking lots to be predicted and the historical remaining space data of various parking lots to be predicted, various parking lots to be predicted are divided into time periods, and the optimal time period division method for various parking lots to be predicted is obtained.

2. The method according to claim 1, characterized in that The cluster reference attributes include one or more of spatial location information, functional area attributes, and historical residual data; The clustering of the parking lots to be predicted based on the clustering reference attributes of the parking lots includes: In the case where the cluster reference attribute includes spatial location information, clustering the parking lots to be predicted according to spatial accessibility and spatial distance of the parking lots to be predicted; wherein the parking lots to be predicted that are spatially accessible and whose spatial distances are within a first distance threshold are clustered into the same category; In the case where the cluster reference attribute includes a functional area attribute, clustering the parking lots to be predicted according to the functional area attributes of the parking lots to be predicted; wherein the parking lots to be predicted with the same functional area attributes are clustered into the same category; In the case where the clustering reference attribute includes historical remaining space data, the morphological similarity of the historical remaining space data of each parking lot to be predicted is determined based on the historical remaining space data of the parking lot to be predicted, and the parking lots to be predicted are clustered based on the morphological similarity of the historical remaining space data of each parking lot to be predicted.

3. The method according to claim 1 or 2, characterized in that: The cluster reference attributes include spatial location information, functional area attributes, and historical residual data; The parking lots to be predicted are clustered according to the clustering reference attributes of the parking lots to obtain k types of parking lots to be predicted, including: Clustering the parking lots to be predicted according to their spatial accessibility and spatial distance to obtain n types of parking lots to be predicted; n≥2; According to the functional area attributes of each parking lot to be predicted, the n types of parking lots to be predicted are clustered again to obtain m types of parking lots to be predicted; wherein m≤n; According to the historical remaining space data of each parking lot to be predicted, the morphological similarity of the historical remaining space data of each parking lot to be predicted in each type of parking lot to be predicted in the m types of parking lots to be predicted is determined respectively, and according to the morphological similarity of the historical remaining space data of each parking lot to be predicted in each type of parking lot to be predicted, each parking lot to be predicted in each type of parking lot to be predicted is clustered to obtain k types of parking lots to be predicted; k≥m.

4. The method according to claim 1, characterized in that: The inter-class differences between the various types of parking lots to be predicted are determined based on the functional area attributes of the various types of parking lots to be predicted; The method of dividing the various parking lots to be predicted into time periods according to the inter-class differences between the various parking lots to be predicted and the historical remaining space data of the various parking lots to be predicted to obtain the optimal time period division method for the various parking lots to be predicted includes: According to the functional area attributes of various parking lots to be predicted, the morphological fluctuations of the historical remaining space data of various parking lots to be predicted, and the initial time period division method of various parking lots to be predicted, the regression model is used to determine the optimal time period division method of various parking lots to be predicted.

5. The method according to claim 4, characterized in that The optimal time period division method for each parking lot to be predicted is determined by using a regression model based on the functional area attributes of each parking lot to be predicted, the morphological fluctuations of the historical remaining space data of each parking lot to be predicted, and the initial time period division method for each parking lot to be predicted, including: Input the functional area attributes of each type of parking lot to be predicted and the initial time period division method of each type of parking lot to be predicted into the tree model to obtain the recommended time period division method of each type of parking lot to be predicted; Using the regression model, the effectiveness of the current recommended time period division method for each type of parking lot to be predicted is evaluated; When the effect evaluation result does not meet the requirements, the functional area attributes of each type of parking lot to be predicted and the current recommended time period division method of each type of parking lot to be predicted are input into the tree model to obtain a new recommended time period division method of each type of parking lot to be predicted; When the effect evaluation results meet the requirements, the currently recommended time period division method for each type of parking lot to be predicted is determined as the optimal time period division method for each type of parking lot to be predicted.

6. The method according to claim 1, characterized in that The method of training a remaining space prediction model for each target time period of each type of parking lot to be predicted based on the historical remaining space data of each target time period of the parking lot to be predicted includes: For any target time period of any type of parking lot to be predicted, a remaining space prediction model corresponding to the target time period of the type of parking lot to be predicted is trained based on the training data of the target time period of the type of parking lot to be predicted; The training data of the target time period of the parking lot to be predicted includes: historical remaining space data of the parking lot to be predicted, and reference remaining space prediction data of adjacent time periods.

7. The method according to claim 6, characterized in that The reference residual prediction data of the adjacent time periods include one or more of the following: The remaining space prediction results of the upstream time period of the parking lot to be predicted of this type, and the remaining space prediction results of the upstream time period of other types of parking lots to be predicted that are most adjacent to the parking lot to be predicted of this type; The most adjacent other types of parking lots to be predicted refers to other types of parking lots to be predicted whose spatial distance to the parking lot to be predicted is less than the second distance threshold and whose spatial distance to the parking lot to be predicted is the shortest.

8. The method according to claim 1, characterized in that The method further comprises: Obtain historical remaining space data of the parking lot to be predicted, and aggregate and count the historical remaining space data of each parking lot to be predicted according to a preset statistical period; For any parking lot to be predicted, if there is data missing within the specified working time range, and the ratio of the duration of data missing to the preset statistical period does not exceed the preset threshold, the missing data is supplemented based on the remaining data of the same period in history; For any parking lot to be predicted, when there are data anomalies in the aggregated statistical results of the parking lot to be predicted, the abnormal data are corrected based on the remaining data of the adjacent statistical period and the remaining data of the same historical period.

9. A parking lot management system, characterized in that: include: A sensor, a memory, and a processor; wherein: The sensor is used to detect vehicles entering and exiting the parking lot, and determine the number of remaining parking spaces in the parking lot based on the detection result; The memory is used to store the remaining space data of the parking lot and the trained remaining space prediction model; wherein the trained remaining space prediction model is trained using the method described in any one of claims 1 to 8; The processor is used to run the trained remaining space prediction model, and use the trained remaining space prediction model to predict the remaining space of the parking lot according to the historical remaining space data of the parking lot stored in the memory to obtain a remaining space prediction result.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor is used to execute the machine executable instructions to implement the method according to any one of claims 1 to 8.

Images