A site spatiotemporal prediction method and system fusing city semantics and representation learning

Abstract

This application discloses a spatiotemporal prediction method and system for charging station sites that integrates urban semantics and representation learning. The spatiotemporal prediction method for charging station sites that integrates urban semantics and representation learning includes: acquiring first historical order data of the site to be predicted and first surrounding point of interest data within a preset distance range of the site to be predicted; determining a first historical time series of the site to be predicted based on the first historical order data; determining first site environment data of the site to be predicted based on the first surrounding point of interest data; inputting the first historical time series and the first site environment data into a trained spatiotemporal fusion model to obtain a first fusion feature vector output by the trained spatiotemporal fusion model; and determining the prediction result of the site to be predicted based on the first fusion feature vector, thereby improving the accuracy of charging station site prediction.

Description

Technical Field

[0001] This application relates to the technical field of site spatiotemporal prediction that integrates urban semantics and representation learning, and in particular to a site spatiotemporal prediction method and system that integrates urban semantics and representation learning. Background Technology

[0002] Current charging station prediction methods mainly rely on pure time-series prediction models such as LSTM, which only rely on the station's own historical order time series for prediction. They ignore the spatial environmental influence factors of the surrounding points of interest and cannot characterize the demand differences brought about by different location environments. The prediction deviation is large in different types of stations such as business districts, residential areas, and suburbs. Alternatively, they simply integrate historical order data and surrounding points of interest data, and directly output the prediction result after simply splicing the two types of features, resulting in low accuracy of the final prediction result. Summary of the Invention

[0003] This application aims to at least address the technical problems existing in the prior art. To this end, this application proposes a spatiotemporal prediction method and system for charging station sites that integrates urban semantics and representation learning, which can improve the accuracy of charging station site prediction.

[0004] The first aspect of this application provides a spatiotemporal prediction method for sites that integrates urban semantics and representation learning, comprising the following steps: Obtain the first historical order data of the site to be predicted and the first surrounding points of interest data within a preset distance range of the site to be predicted; Based on the first historical order data, a first historical time series of the site to be predicted is determined; Based on the first surrounding point of interest data, determine the first site environment data of the site to be predicted; The first historical time series and the first site environmental data are input into the trained spatiotemporal fusion model to obtain the first fusion feature vector output by the trained spatiotemporal fusion model. Based on the first fused feature vector, the prediction result of the site to be predicted is determined.

[0005] The site spatiotemporal prediction method integrating urban semantics and representation learning according to the embodiments of this application has at least the following beneficial effects: This application acquires historical order data and surrounding point-of-interest data for the site to be predicted, transforms them into historical time series and site environment data, and then inputs them into a spatiotemporal fusion model to generate fusion features, ultimately determining the prediction result. Therefore, this application can effectively utilize the functional semantic information of the city surrounding the charging station, distinguish differences in charging behavior under different scenarios, and achieve deep fusion of heterogeneous modal information, providing a more accurate data basis for the prediction model, thereby improving the accuracy of charging station site prediction.

[0006] A second aspect of this application provides a site spatiotemporal prediction system that integrates urban semantics and representation learning. The site spatiotemporal prediction system that integrates urban semantics and representation learning includes: The data acquisition module is used to acquire the first historical order data of the site to be predicted and the first surrounding points of interest data within a preset distance range of the site to be predicted; The first historical time series determination module is used to determine the first historical time series of the site to be predicted based on the first historical order data. The first site environment data determination module is used to determine the first site environment data of the site to be predicted based on the first surrounding point of interest data. The spatiotemporal fusion module is used to input the first historical time series and the first site environmental data into the trained spatiotemporal fusion model to obtain the first fusion feature vector output by the trained spatiotemporal fusion model. The prediction module is used to determine the prediction result of the site to be predicted based on the first fused feature vector.

[0007] This system acquires historical order data and surrounding point-of-interest data for the charging station to be predicted, transforms them into historical time series and station environment data, and then inputs them into a spatiotemporal fusion model to generate fusion features, ultimately determining the prediction result. Therefore, this application can effectively utilize the functional semantic information of the city surrounding the charging station, distinguish differences in charging behavior under different scenarios, and achieve deep fusion of heterogeneous modal information, providing a more accurate data basis for the prediction model, thereby improving the accuracy of charging station prediction.

[0008] A third aspect of this application provides an electronic device including at least one processor and a memory for communicatively connecting to the processor; the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform a site spatiotemporal prediction method integrating urban semantics and representation learning as described in the first aspect of this application.

[0009] A fourth aspect of this application provides a computer-readable storage medium storing computer-executable instructions for causing a computer to perform a site spatiotemporal prediction method integrating urban semantics and representation learning as described in the first aspect of this application.

[0010] It should be noted that the beneficial effects of the third and fourth aspects of this application compared with the prior art are the same as the beneficial effects of the above-mentioned site spatiotemporal prediction method integrating urban semantics and representation learning compared with the prior art, and will not be elaborated here.

[0011] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0012] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart illustrating an embodiment of the site spatiotemporal prediction method that integrates urban semantics and representation learning provided in this application. Figure 2 This is a schematic diagram of the structure of an embodiment of the site spatiotemporal prediction system that integrates urban semantics and representation learning provided in this application; Figure 3 This is a schematic diagram of the structure of an embodiment of the electronic device provided in this application. Detailed Implementation

[0013] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0014] In the description of this application, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated.

[0015] In the description of this application, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0016] In the description of this application, it should be noted that, unless otherwise explicitly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution.

[0017] Current charging station site prediction methods mainly rely on pure time-series prediction models such as LSTM, which only rely on the site's own historical order time series for prediction. They ignore the spatial environmental influence factors of the surrounding points of interest, and cannot characterize the demand differences brought about by different location environments. The prediction deviation is large in different types of sites such as business districts, residential areas, and suburbs. Alternatively, they simply merge historical order and surrounding point of interest data, and directly output the prediction result after simply splicing the two types of features. This results in low accuracy of the final prediction result and cannot provide reliable data support for the resource scheduling of charging station sites.

[0018] To address the aforementioned technical deficiencies, this application provides a site spatiotemporal prediction method and system that integrates urban semantics and representation learning.

[0019] Please see Figure 1 This is a flowchart illustrating a site spatiotemporal prediction method integrating urban semantics and representation learning, provided in an embodiment of this application. This method is applied to an electronic device, which may be a server, etc. Figure 1 As shown, this site spatiotemporal prediction method that integrates urban semantics and representation learning includes: Step S101: Obtain the first historical order data of the site to be predicted and the first surrounding points of interest data within a preset distance range of the site to be predicted; The above-mentioned sites can be charging stations for electric vehicles.

[0020] The above-mentioned preset distance range can be a value that can be preset according to actual needs.

[0021] The aforementioned Point of Interest (POI) can be a discrete geographic entity location on a map with a unique identifier.

[0022] Step S102: Based on the first historical order data, determine the first historical time series of the site to be predicted; Step S103: Based on the first surrounding point of interest data, determine the first site environment data of the site to be predicted; Step S104: Input the first historical time series and the first site environmental data into the trained spatiotemporal fusion model to obtain the first fusion feature vector output by the trained spatiotemporal fusion model; Step S105: Based on the first fused feature vector, determine the prediction result of the site to be predicted.

[0023] The above prediction results can be at least one of the following: the occupancy time of several preset time steps in the future, the total charging power, and the total number of orders. The preset time step can be a value set in advance according to actual needs, and can be 1 hour.

[0024] This application acquires historical order data and surrounding point-of-interest data for the site to be predicted, transforms them into historical time series and site environment data, and then inputs them into a spatiotemporal fusion model to generate fusion features, ultimately determining the prediction result. Therefore, this application can effectively utilize the functional semantic information of the city surrounding the charging station, distinguish differences in charging behavior under different scenarios, and achieve deep fusion of heterogeneous modal information, providing a more accurate data basis for the prediction model, thereby improving the accuracy of charging station site prediction.

[0025] In some embodiments, steps S201 to S202 may be included before step S101: Step S201: Obtain the order start time and the order duration of each historical order for the site to be predicted; In step S201, obtaining the order start time and order duration of each historical order of the site to be predicted can be achieved by parsing the original transaction records in the charging station management system, extracting the start timestamp and end timestamp of each historical order of the site to be predicted, using the start timestamp of each historical order as the order start time of each historical order, calculating the difference between the end timestamp and the start timestamp, and using the difference as the order duration of the corresponding historical order.

[0026] Step S202: After obtaining the first location coordinates of the site to be predicted, determine the interest point data within a preset distance range of the site to be predicted based on the first location coordinates. The interest point data includes the second location coordinates of each interest point and the category of each interest point. The first location coordinates mentioned above can be the longitude and latitude of the station to be predicted.

[0027] The second location coordinates mentioned above can be the longitude and latitude of the point of interest.

[0028] The above categories can be residential, office, commercial, transportation, medical, or educational.

[0029] In step S202, the above-mentioned determination of the point of interest data within a preset distance range of the site to be predicted based on the first location coordinates can be achieved by calling the map service interface, setting a preset distance range with the first location coordinates of the site to be predicted as the center, querying all points of interest within the preset distance range, extracting the second location coordinates of each point of interest, and manually labeling the category of each point of interest.

[0030] Step S101 may include steps S203 to S204: Step S203: Use the order start time and order duration as the first historical order data; Step S204: Use the point of interest data as the first surrounding point of interest data.

[0031] This application first transforms the original discrete order records into sequence data with a time span attribute by obtaining the order start time and duration of each historical order. This transformation allows the historical order data to more accurately reflect the occupancy characteristics and load changes of the site in different time periods, solving the problem that the number of orders alone cannot characterize the continuity of charging behavior. This lays the foundation for subsequently constructing a refined first historical time series. Then, by obtaining the first location coordinates of the site to be predicted and determining the point of interest data within a preset distance range centered on these coordinates, spatial filtering of the surrounding point of interest data is achieved. This ensures that the obtained point of interest data has a real geographical correlation with the site and avoids interference from irrelevant noise data. The second location coordinates and category information contained in the point of interest data provide necessary multi-dimensional feature support for subsequent analysis of the functional semantics around the site. These preprocessed and structured order start times, order durations, and point of interest data are then used as the first historical order data and the first surrounding point of interest data, respectively. This results in higher quality and stronger discriminative power in the data input to the spatiotemporal fusion model. This data preprocessing method effectively improves the model's ability to capture the dynamic patterns of charging behavior from a temporal dimension and perceive the urban functional attributes around the site from a spatial dimension. This significantly enhances the spatiotemporal fusion model's ability to represent the spatiotemporal characteristics of the site, thereby improving the accuracy and reliability of the prediction results.

[0032] In some embodiments, step S102 may include steps S301 to S302: Step S301: Based on the order start time and order duration, determine the occupancy time, total charging power, and total number of orders of the site to be predicted within the preset time step; In step S301, determining the occupancy time, total charging capacity, and total number of orders of the site to be predicted within a preset time step based on the order start time and order duration may include: Based on the order start time and order duration, the duration of the site to be predicted within the preset time step is calculated using the following formula: ; ; in, For the site to be predicted in the 1st The duration of each preset time step. The first parameter value is preset according to actual needs. This is a second parameter value pre-set according to actual needs. This is a third parameter value pre-set according to actual needs. For the site to be predicted in the 1st The sum of the duration of all historical orders corresponding to a preset time step. The site to be predicted The duration of the first historical order and the order duration of the second order The overlap duration of a preset time step For the first The start time of each historical order. For the first The duration of each historical order. For the first The end time of a preset time step. For the first The start time of a preset time step; Based on the order start time and order duration, the total charging capacity of the site to be predicted within a preset time step is calculated using the following formula: ; in, For the site to be predicted in the 1st The total charging capacity for a preset time step. For the pre-acquired first The charging amount corresponding to each historical order; Based on the order start time and order duration, the total number of orders for the site to be predicted within a preset time step is calculated using the following formula: ; in, For the site to be predicted in the 1st Total number of orders within a preset time step This is an indicator function.

[0033] Step S302: Construct the first historical time series based on the duration of occupation, total charging power, and total number of orders.

[0034] In step S302, the construction of the first historical time series based on the occupancy time, total charging power, and total number of orders can be achieved by taking the occupancy time, total charging power, and total number of orders within each preset time step as a feature vector of that time step, and then concatenating these feature vectors in chronological order to obtain the first historical time series.

[0035] This application transforms discrete order start times and order durations into occupancy duration, total charging power, and total number of orders within a preset time step, achieving a refined characterization of the dynamic operating status of a site. By using the first historical time series, it can comprehensively perceive the spatiotemporal evolution characteristics of the site to be predicted from three complementary perspectives: occupancy, load, and frequency. This provides more representative and physically meaningful input data for subsequent spatiotemporal fusion models, thereby significantly improving the accuracy of the model in capturing the spatiotemporal evolution patterns of sites, and thus enhancing the accuracy and robustness of the prediction results.

[0036] In some embodiments, step S103 may include steps S401 to S403: Step S401: Based on the first and second location coordinates, determine the distance between each point of interest and the station to be predicted; In step S401, the above-mentioned determination of the distance between each point of interest and the station to be predicted based on the first location coordinates and the second location coordinates can be achieved by calculating the Haversine distance between the first location coordinates and the second location coordinates, and using the Haversine distance as the distance between the corresponding point of interest and the station to be predicted.

[0037] Step S402: Determine the number of categories for each category in the first surrounding point of interest data; In step S402, the number of categories in the first surrounding interest point data can be obtained by traversing the first surrounding interest point data and accumulating the number of interest points under each category.

[0038] Step S403: Based on the number of categories and distance, determine the first site environment data of the site to be predicted.

[0039] In step S403, the first site environment data for determining the site to be predicted based on the number of categories and distance may include: Divide the number of categories by the total number of all categories to obtain the category ratio for each category. The minimum distance between all points of interest in each category and the site to be predicted is selected as the nearest distance value for each category. The distance decay proximity for each category is calculated using the following formula: ; in, For the first The similarity of distance decay corresponding to each category For the first The first category The distance between each point of interest and the site to be predicted These are preset distance attenuation parameter values ​​set in advance according to actual needs.

[0040] The number of categories, category ratio, nearest distance value, and distance decay proximity of each category of the site to be predicted are used as the first site environment data of the site to be predicted.

[0041] This application effectively distinguishes the influence weights of interest points on the functional attributes of a site at different spatial scales by calculating the distance between each interest point and the site to be predicted. This avoids semantic ambiguity caused by treating all surrounding interest points as equally important. Secondly, based on the statistical count of the number of interest points of each category, a comprehensive calculation is performed by combining distance to construct a vector or numerical value that reflects the functional distribution characteristics around the site. This allows for a more accurate characterization of the type of urban functional area where the site is located. This approach enables the model to perceive the differences in the impact of functional facilities on charging demand within different distance ranges, providing more discriminative and representative input features for subsequent spatiotemporal fusion models. It effectively solves the problem that a single statistical method is difficult to capture the real spatial distribution patterns of urban functions, thereby improving the spatiotemporal fusion model's ability to perceive and predict differences in site charging behavior.

[0042] In some embodiments, the training process of the trained spatiotemporal fusion model may include steps S501 to S506: Step S501: Determine the second historical time series and second site environmental data for each site; In step S501, the calculation process for determining the second historical time series of each station is similar to the calculation process for determining the first historical time series of the station to be predicted in step S102, and will not be repeated here.

[0043] In step S501, the calculation process for determining the second site environmental data for each site is similar to the calculation process for determining the first site environmental data for the site to be predicted in step S103, and will not be repeated here.

[0044] Step S502: Construct an initial spatiotemporal fusion model, wherein the initial spatiotemporal fusion model includes an initial time feature encoding module, an initial city context encoding module, and an initial spatiotemporal fusion module; The aforementioned initial spatiotemporal fusion model can be an untrained deep learning model.

[0045] Step S503: Input the second historical time series into the initial time feature encoding module to obtain the time feature vector output by the initial time feature encoding module; In step S503, the above-mentioned input of the second historical time series into the initial time feature encoding module to obtain the time feature vector output by the initial time feature encoding module can be: inputting the second historical time series into the initial time feature encoding module to extract features through the one-dimensional causal dilated convolutional network of the initial time feature encoding module, obtaining the feature extraction result, pooling the feature extraction result through the pooling layer of the initial time feature encoding module, obtaining the pooling result, and inputting the pooling result into the fully connected layer of the initial time feature encoding module to obtain the time feature vector output by the fully connected layer.

[0046] Step S504: Input the second site environmental data into the initial city context coding module to obtain the city context feature vector output by the initial city context coding module; The initial city context encoding module mentioned above can be a multilayer perceptron module.

[0047] In step S504, the above-mentioned input of the second site environmental data into the initial urban context encoding module to obtain the urban context feature vector output by the initial urban context encoding module can be achieved by inputting the second site environmental data into the multilayer perceptron module for nonlinear transformation to obtain the urban context feature vector.

[0048] Step S505: Input the time feature vector and the city context feature vector into the initial spatiotemporal fusion module to obtain the second fused feature vector output by the initial spatiotemporal fusion module; The aforementioned initial spatiotemporal fusion module can be a gated fusion module.

[0049] In step S505, the above-mentioned inputting the time feature vector and the city context feature vector into the initial spatiotemporal fusion module to obtain the second fused feature vector output by the initial spatiotemporal fusion module can be achieved by inputting the time feature vector and the city context feature vector into the gating fusion module for feature fusion to obtain the second fused feature vector.

[0050] Step S506: Based on the second historical time series and the second fusion feature vector, iteratively update the initial spatiotemporal fusion model to obtain the trained spatiotemporal fusion model.

[0051] This application provides the model with dual inputs—including dynamic behavioral patterns and static spatial semantics—by determining the second historical time series and environmental data of the site. The initial time feature encoding module processes the time series, extracting the periodic and trend features of the site's charging behavior. The initial urban context encoding module processes the environmental data, transforming the distribution of surrounding points of interest into feature vectors reflecting urban functional semantics. A spatiotemporal fusion module then maps both to a unified feature space, generating a second fused feature vector, thus achieving information interaction between the time and spatial semantic dimensions. Finally, an iterative update mechanism enables the model to learn more robust feature representations, effectively addressing the problem of insufficient utilization of single-modal information and strengthening the model's ability to capture spatiotemporal correlation patterns, thereby improving the model's predictive generalization performance in complex urban environments.

[0052] In some embodiments, step S506 may include steps S601 to S604: Step S601: Based on the second historical time series and the second fused feature vector, determine the reconstruction loss value through the reconstruction loss function; The reconstruction loss function mentioned above can be the mean squared error function or the L2 loss function.

[0053] In step S601, the reconstruction loss value is determined by the reconstruction loss function based on the second historical time series and the second fused feature vector. This can be achieved by reconstructing the second fused feature vector through a decoder (such as a fully connected layer or a deconvolution layer) to obtain the reconstructed time series. The mean square error (MSE) between the reconstructed time series and the time feature vector of the second historical time series is calculated using the mean square error function, and the MSE is used as the reconstruction loss value.

[0054] Step S602: Based on the second fused feature vector, determine the contrast loss value through the contrast loss function; The contrast loss function mentioned above can be the InfoNCE loss function.

[0055] In step S602, based on the second fusion feature vector, the comparison loss value is determined by the comparison loss function to construct positive sample pairs (it should be noted that positive sample pairs can be fusion features of different preset time steps of the same site) and negative sample pairs (it should be noted that negative sample pairs can be fusion features of different sites). The InfoNCE loss value of the positive sample pairs and negative sample pairs is calculated by the InfoNCE loss function, and the InfoNCE loss value is used as the comparison loss value.

[0056] Step S603: Determine the total loss value based on the reconstruction loss value and the comparison loss value; In step S603, determining the total loss value based on the reconstruction loss value and the contrast loss value may include: The reconstruction loss value is multiplied by the first preset weight value pre-set according to actual needs to obtain the first product value; The comparison loss value is multiplied by a second preset weight value pre-set according to actual needs to obtain the second product value; Add the first product value to the second product value to get the total loss value.

[0057] The sum of the first preset weight value and the second preset weight value is one.

[0058] Step S604: If the total loss value is less than the preset loss threshold, use the initial spatiotemporal fusion model as the trained spatiotemporal fusion model.

[0059] The aforementioned preset loss threshold can be a value set in advance according to actual needs.

[0060] In some embodiments, the method may further include: If the total loss value is greater than or equal to the preset loss threshold, the parameters of the initial spatiotemporal fusion model are iteratively updated according to the total loss value through the backpropagation algorithm until the number of iterations reaches the maximum number of iterations preset according to actual needs, and the trained spatiotemporal fusion model is obtained.

[0061] This application enables the model to effectively recover the original temporal feature information from the fused features by reconstructing the loss function. This process forces the model to retain key time series dynamic patterns during fusion, preventing information loss during feature extraction. Then, by using a contrastive loss function, the model performs discriminative learning on the fused features in the feature space. This helps enhance the model's ability to distinguish between different site environments and behavioral patterns, ensuring that the generated fused features not only contain temporal information but also deeply integrate the differences in urban functional semantics. Furthermore, by weighting or combining the reconstruction loss and contrastive loss into a total loss value, a dual constraint on the model's representation ability is achieved, ensuring both the integrity of the features and improving their generalization and discriminativeness. Finally, iteration stops when the total loss value is lower than a preset threshold, ensuring that the model completes training after reaching the expected feature representation accuracy. This provides high-quality input features for subsequent site prediction tasks, effectively solving the problems of insufficient heterogeneous modality fusion and insufficient representation generalization ability, and significantly improving the model's prediction accuracy and robustness in complex urban scenarios.

[0062] In some embodiments, step S105 may include step S701: Step S701: Input the first fused feature vector into the ridge regression prediction model to obtain the prediction result output by the ridge regression prediction model.

[0063] This application obtains the prediction results output by inputting the first fused feature vector into the ridge regression prediction model, which can more robustly capture the intrinsic relationship between urban semantic information and charging behavior, thereby improving the accuracy and generalization ability of the prediction results.

[0064] Specifically, refer to Table 1, which is a comparison table of prediction performance for usage time. In Table 1, Changsha represents the experimental dataset constructed using daily usage data of charging stations in Changsha; Dundee represents the experimental dataset constructed using daily usage data of charging stations in Dundee; RMSE represents the root mean square error; WAPE represents the weighted absolute percentage error; R² represents the coefficient of determination; LSTM represents the existing long short-term memory network model; Transformer represents the existing deep learning model based on self-attention mechanism; ARIMA represents the existing autoregressive integral moving average model; TimePFN represents an existing pre-trained basic model for time series prediction; SimpleTM represents an existing time series prediction model; ST-SSDL represents an existing spatiotemporal self-supervised deviation learning model; ST-ReP represents an existing spatiotemporal representation learning model; ChatTime represents an existing prediction model for time series tasks; and Time-LLM represents an existing time series prediction model based on a large language model.

[0065] Table 1

[0066] As shown in Table 1, compared with LSTM, Transformer, ARIMA, TimePFN, SimpleTM, ST-ReP, ST-SSDL, ChatTime, and Time-LLM, this application achieves lower RMSE and WAPE and higher R² in the occupancy duration prediction task, indicating that this application can effectively improve the prediction performance of occupancy duration.

[0067] Referring to Table 2, which is a comparison table of prediction performance for total charging capacity, it can be seen from Table 2 that compared with LSTM, Transformer, ARIMA, TimePFN, SimpleTM, ST-ReP, ST-SSDL, ChatTime and Time-LLM, this application achieves lower RMSE and WAPE and higher R² in the prediction task of total charging capacity, indicating that this application can effectively improve the prediction performance of total charging capacity.

[0068] Table 2

[0069] Referring to Table 3, which shows the results of the ablation experiment, "Removing urban context" means that the site environment data of this application is not used, and prediction is made only based on historical time series; "Removing time encoding" means that the historical time series of this application is not used, and prediction is made only based on site environment data; "Removing contrast loss" means that the contrast loss value is not calculated during training, and backpropagation optimization is performed only based on the reconstruction loss value; "Complete model" means the complete method flow of this application.

[0070] Table 3

[0071] As shown in Table 3, compared with each ablation variant, the complete model performs better in predicting the duration of occupancy and the total charging capacity on the Changsha and Dundee datasets.

[0072] Additionally, refer to Figure 2 One embodiment of this application provides a site spatiotemporal prediction system that integrates urban semantics and representation learning, including a data acquisition module 1100, a first historical time series determination module 1200, a first site environment data determination module 1300, a spatiotemporal fusion module 1400, and a prediction module 1500, wherein: The data acquisition module 1100 is used to acquire the first historical order data of the site to be predicted and the first surrounding points of interest data within a preset distance range of the site to be predicted; The first historical time series determination module 1200 is used to determine the first historical time series of the site to be predicted based on the first historical order data; The first site environment data determination module 1300 is used to determine the first site environment data of the site to be predicted based on the first surrounding point of interest data. The spatiotemporal fusion module 1400 is used to input the first historical time series and the first site environmental data into the trained spatiotemporal fusion model to obtain the first fusion feature vector output by the trained spatiotemporal fusion model; The prediction module 1500 is used to determine the prediction result of the site to be predicted based on the first fused feature vector.

[0073] This system acquires historical order data and surrounding point-of-interest data for the charging station to be predicted, transforms them into historical time series and station environment data, and then inputs them into a spatiotemporal fusion model to generate fusion features, ultimately determining the prediction result. Therefore, this application can effectively utilize the functional semantic information of the city surrounding the charging station, distinguish differences in charging behavior under different scenarios, and achieve deep fusion of heterogeneous modal information, providing a more accurate data basis for the prediction model, thereby improving the accuracy of charging station prediction.

[0074] It should be noted that the system embodiments described above are based on the same inventive concept as the method embodiments described above. Therefore, the relevant content of the method embodiments described above is also applicable to the system embodiments described above, and will not be repeated here.

[0075] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations. The acquisition, storage, use and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations.

[0076] like Figure 3 One embodiment of this application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned method for spatiotemporal prediction of sites that integrates urban semantics and representation learning. The electronic device includes: At least one memory; At least one processor; At least one program; The program is stored in memory, and the processor executes at least one program to implement a site spatiotemporal prediction method that integrates urban semantics and representation learning according to the above embodiments of this disclosure.

[0077] Electronic devices can be any smart terminal, including mobile phones, tablets, personal digital assistants (PDAs), and in-vehicle computers.

[0078] The electronic devices according to embodiments of this application will now be described in detail.

[0079] The processor 1600 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure. The memory 1700 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1700 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1700 and is called and executed by the processor 1600 to execute a site spatiotemporal prediction method that integrates urban semantics and representation learning according to an embodiment of this disclosure.

[0080] The input / output interface 1800 is used to implement information input and output. The communication interface 1900 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 2000 transmits information between various components of the device (e.g., processor 1600, memory 1700, input / output interface 1800, and communication interface 1900); The processor 1600, memory 1700, input / output interface 1800 and communication interface 1900 are connected to each other within the device via bus 2000.

[0081] This disclosure also provides a storage medium, which is a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the above-described site spatiotemporal prediction method that integrates urban semantics and representation learning.

[0082] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0083] The above is a detailed description of the preferred embodiments of this application. However, the embodiments of this application are not limited to the above-described implementation methods. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the embodiments of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of the embodiments of this application.

Claims

1. A method for spatio-temporal prediction of sites fusing urban semantics and representation learning, characterized in that, The site spatiotemporal prediction method that integrates urban semantics and representation learning includes: Obtain the first historical order data of the site to be predicted and the first surrounding points of interest data within a preset distance range of the site to be predicted; Based on the first historical order data, a first historical time series of the site to be predicted is determined; Based on the first surrounding point of interest data, determine the first site environment data of the site to be predicted; The first historical time series and the first site environmental data are input into the trained spatiotemporal fusion model to obtain the first fusion feature vector output by the trained spatiotemporal fusion model. Based on the first fused feature vector, the prediction result of the site to be predicted is determined. 2.The method of claim 1, wherein, Before acquiring the first historical order data of the site to be predicted and the first surrounding point of interest data within a preset distance range of the site to be predicted, the process includes: Obtain the order start time and the order duration of each historical order for the site to be predicted; In the case of obtaining the first location coordinates of the site to be predicted, interest point data within the preset distance range of the site to be predicted is determined based on the first location coordinates, wherein the interest point data includes the second location coordinates of each interest point and the category of each interest point; The acquisition of the first historical order data of the site to be predicted and the first surrounding points of interest data within a preset distance range of the site to be predicted includes: The order start time and the order duration are used as the first historical order data; The point of interest data is used as the first surrounding point of interest data. 3.The method of claim 2, wherein, The step of determining the first historical time series of the site to be predicted based on the first historical order data includes: Based on the order start time and the order duration, determine the occupancy time, total charging power, and total number of orders of the site to be predicted within a preset time step; The first historical time series is constructed based on the occupied time, the total charging power, and the total number of orders.

4. The method of claim 2, wherein the method further comprises: The step of determining the first site environment data of the site to be predicted based on the first surrounding point of interest data includes: Based on the first location coordinates and the second location coordinates, the distance between each point of interest and the site to be predicted is determined; Determine the number of categories for each category in the first surrounding point of interest data; Based on the number of categories and the distance, the first site environment data of the site to be predicted is determined.

5. The method of claim 1, wherein, The training process of the trained spatiotemporal fusion model includes: Determine the second historical time series and second site environmental data for each site; An initial spatiotemporal fusion model is constructed, wherein the initial spatiotemporal fusion model includes an initial time feature encoding module, an initial city context encoding module, and an initial spatiotemporal fusion module; The second historical time series is input into the initial time feature encoding module to obtain the time feature vector output by the initial time feature encoding module; The second site environment data is input into the initial city context encoding module to obtain the city context feature vector output by the initial city context encoding module; The time feature vector and the city context feature vector are input into the initial spatiotemporal fusion module to obtain the second fused feature vector output by the initial spatiotemporal fusion module; Based on the second historical time series and the second fusion feature vector, the initial spatiotemporal fusion model is iteratively updated to obtain the trained spatiotemporal fusion model.

6. The method of claim 5, wherein the method further comprises: The step of iteratively updating the initial spatiotemporal fusion model based on the second historical time series and the second fusion feature vector to obtain the trained spatiotemporal fusion model includes: Based on the second historical time series and the second fused feature vector, the reconstruction loss value is determined by the reconstruction loss function; Based on the second fused feature vector, the contrastive loss value is determined by the contrastive loss function; Based on the reconstruction loss value and the comparison loss value, the total loss value is determined; If the total loss value is less than the preset loss threshold, the initial spatiotemporal fusion model is used as the trained spatiotemporal fusion model.

7. The method of claim 1, wherein, The step of determining the prediction result of the site to be predicted based on the first fused feature vector includes: The first fused feature vector is input into the ridge regression prediction model to obtain the prediction result output by the ridge regression prediction model.

8. A site spatio-temporal prediction system that fuses urban semantics with representation learning, characterized in that, The site spatiotemporal prediction system that integrates urban semantics and representation learning includes: The data acquisition module is used to acquire the first historical order data of the site to be predicted and the first surrounding points of interest data within a preset distance range of the site to be predicted; The first historical time series determination module is used to determine the first historical time series of the site to be predicted based on the first historical order data. The first site environment data determination module is used to determine the first site environment data of the site to be predicted based on the first surrounding point of interest data. The spatiotemporal fusion module is used to input the first historical time series and the first site environmental data into the trained spatiotemporal fusion model to obtain the first fusion feature vector output by the trained spatiotemporal fusion model. The prediction module is used to determine the prediction result of the site to be predicted based on the first fused feature vector.

9. An electronic device, comprising: It includes at least one processor and a memory for communicatively connecting to the processor; the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform a site spatiotemporal prediction method that integrates urban semantics and representation learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions for causing a computer to perform a site spatiotemporal prediction method that integrates urban semantics and representation learning as described in any one of claims 1 to 7.

Images