Charging pile occupancy rate prediction method and device, processor and electronic equipment

Abstract

The invention discloses a charging pile occupancy rate prediction method and device, a processor and electronic equipment. The method relates to the technical field of electric vehicles, and comprises the steps that a plurality of time-space sub-graphs corresponding to a plurality of charging piles in an area are acquired, and one time-space sub-graph is used for indicating at least two pieces of state data of one charging pile; under the condition that at least two time-space sub-graphs meeting an expected condition are determined from the multiple time-space sub-graphs, the at least two time-space sub-graphs are combined to obtain a time-space sub-graph cluster, and the expected condition is used for indicating that the similarity degree of the at least two time-space sub-graphs is larger than a preset threshold value; feature reconstruction is carried out on each time-space sub-graph in the time-space sub-graph cluster, and the reconstructed time-space sub-graphs indicate more state data of the charging pile; and predicting the occupancy rate of the charging piles in the area by using the reconstructed time-space sub-graph cluster.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric vehicles, in particular to a charging pile occupancy rate prediction method and device, a processor and an electronic device. BACKGROUND

[0002] The charging demand fluctuation of charging piles in a charging network has significant spatio-temporal heterogeneity, and each charging pile can include multi-dimensional state data, such as charging power, occupancy state, geographic location coordinates, etc., which makes it difficult to ensure prediction accuracy when dealing with complex and variable data, and further leads to low accuracy of charging pile occupancy rate prediction.

[0003] In view of the low accuracy of charging pile occupancy rate prediction in the related art, no effective solution has been proposed so far. SUMMARY

[0004] The main purpose of the present application is to provide a charging pile occupancy rate prediction method, device, processor and electronic device to solve the problem of low accuracy of charging pile occupancy rate prediction in the related art.

[0005] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a charging pile occupancy rate prediction method is provided. The method comprises: obtaining a plurality of spatio-temporal subgraphs corresponding to a plurality of charging piles in a region, wherein one spatio-temporal subgraph is used to indicate at least two items of state data of one charging pile; in the case that at least two spatio-temporal subgraphs meeting an expected condition are determined from the plurality of spatio-temporal subgraphs, combining the at least two spatio-temporal subgraphs to obtain a spatio-temporal subgraph cluster, wherein the expected condition is used to indicate that the similarity of the at least two spatio-temporal subgraphs is greater than a preset threshold; reconstructing the features of each spatio-temporal subgraph in the spatio-temporal subgraph cluster, wherein the reconstructed spatio-temporal subgraph indicates more items of state data of the charging pile; and predicting the occupancy rate of the charging pile in the region by using the reconstructed spatio-temporal subgraph cluster.

[0006] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a charging pile occupancy rate prediction device is provided. The device comprises: an obtaining unit configured to obtain a plurality of spatio-temporal subgraphs corresponding to a plurality of charging piles in a region, wherein one spatio-temporal subgraph is used to indicate at least two items of state data of one charging pile; a combining unit configured to combine at least two spatio-temporal subgraphs to obtain a spatio-temporal subgraph cluster in the case that the at least two spatio-temporal subgraphs meeting an expected condition are determined from the plurality of spatio-temporal subgraphs, wherein the expected condition is used to indicate that the similarity of the at least two spatio-temporal subgraphs is greater than a preset threshold; a reconstructing unit configured to reconstruct the features of each spatio-temporal subgraph in the spatio-temporal subgraph cluster, wherein the reconstructed spatio-temporal subgraph indicates more items of state data of the charging pile; and a predicting unit configured to predict the occupancy rate of the charging pile in the region by using the reconstructed spatio-temporal subgraph cluster.

[0007] By the present application, the following steps are adopted: obtaining a plurality of spatio-temporal subgraphs corresponding to a plurality of charging piles in a region, wherein one spatio-temporal subgraph is used to indicate at least two items of state data of one charging pile; in the case of determining at least two spatio-temporal subgraphs that meet the expected condition from the plurality of spatio-temporal subgraphs, combining the at least two spatio-temporal subgraphs to obtain a spatio-temporal subgraph cluster, wherein the expected condition is used to indicate that the similarity degree of the at least two spatio-temporal subgraphs is greater than a preset threshold; reconstructing each spatio-temporal subgraph in the spatio-temporal subgraph cluster, wherein the reconstructed spatio-temporal subgraph indicates more items of state data of the charging pile; and using the reconstructed spatio-temporal subgraph cluster to predict the occupancy rate of the charging pile in the region. Obtaining a plurality of spatio-temporal subgraphs corresponding to a plurality of charging piles in a region, each subgraph not only contains the state data of a single charging pile, but also reflects the spatio-temporal correlation between the charging pile and the surrounding environment. By setting the "expected condition" (such as the similarity degree being greater than a preset threshold), the present application can automatically identify and combine subgraphs with high similarity to form a spatio-temporal subgraph cluster. The feature reconstruction of each subgraph in the spatio-temporal subgraph cluster expands the input dimension of each subgraph, so as to capture the networked conduction effect of the charging pile in space and the nonlinear fluctuation characteristics in time. This enhanced feature representation capability enables the model to more accurately understand the coordinated evolution law of the charging demand in the spatio-temporal continuum, solves the problems of computational efficiency and spatio-temporal feature fusion in the occupancy rate prediction of the large-scale electric vehicle charging network, and realizes the substantial improvement of prediction accuracy and the substantial improvement of prediction timeliness, thereby realizing the technical effect of effectively improving the accuracy of the occupancy rate prediction of the charging pile, and solving the technical problem of low accuracy of the occupancy rate prediction of the charging pile in the related art. BRIEF DESCRIPTION OF DRAWINGS

[0008] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of this application illustrated in the drawings, and their description, are presented to explain the present application and are not intended to limit the present application. In the drawings:

[0009] Figure 1 is a flowchart of a charging pile occupancy rate prediction method provided according to an embodiment of the present application;

[0010] Figure 2 is a flowchart of a spatio-temporal synchronous prediction method based on hierarchical clustering and expert mixing provided according to an embodiment of the present application;

[0011] Figure 3 is a reconstructed adjacency matrix diagram provided according to an embodiment of the present application;

[0012] Figure 4 is a schematic diagram of a charging pile occupancy rate prediction device provided according to an embodiment of the present application;

[0013] Figure 5A schematic diagram of an occupancy rate prediction electronic device of a charging pile is provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0014] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0015] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the scope of protection of the present application.

[0016] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product or device.

[0017] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, analyzed data, etc.) involved in the present disclosure are all information and data authorized by the user or authorized by all parties. For example, an interface is provided between the system and the relevant user or institution. Before obtaining the relevant information, the interface needs to send a request to the aforementioned user or institution, and after receiving the consent information fed back by the aforementioned user or institution, the relevant information is obtained.

[0018] The present application will be described below in combination with the preferred implementation steps, Figure 1 A flowchart of a charging pile occupancy rate prediction method is provided according to an embodiment of the present application, as shown in Figure 1 The method comprises the following steps:

[0019] S101, obtaining a plurality of spatio-temporal subgraphs corresponding to a plurality of charging piles in a region, wherein one spatio-temporal subgraph is used to indicate at least two state data of one charging pile;

[0020] S102, in the case of determining at least two spatio-temporal subgraphs meeting the expected condition from the plurality of spatio-temporal subgraphs, combining the at least two spatio-temporal subgraphs to obtain a spatio-temporal subgraph cluster, wherein the expected condition is used to indicate that the similarity of the at least two spatio-temporal subgraphs is greater than a preset threshold;

[0021] S103, reconstructing the features of each spatio-temporal subgraph in the spatio-temporal subgraph cluster, wherein the reconstructed spatio-temporal subgraph indicates more state data of the charging pile;

[0022] S104, using the reconstructed spatio-temporal subgraph cluster to predict the occupancy rate of the charging pile in the region.

[0023] Optionally, in the present embodiment, in an electric vehicle charging network, each charging pile is regarded as a node, and the relationship (such as geographical proximity, similarity of charging behavior) between nodes is represented by an edge, forming a graph structure. The spatio-temporal subgraph is a local subnetwork obtained by dividing the global graph structure through the agglomerative hierarchical clustering algorithm, which contains the state data and spatio-temporal correlation information of the charging pile. The "state data" here includes at least two items, which can be charging power, occupancy state, geographical position coordinates, cumulative charging duration, etc. The spatio-temporal subgraph reflects the dynamic connection and interaction of the charging pile with other charging piles within a certain spatio-temporal range through local aggregation of data.

[0024] Optionally, in the present embodiment, when a plurality of spatio-temporal subgraphs meet certain similarity criteria ("expected condition"), they will be combined into a "spatio-temporal subgraph cluster". The expected condition is usually based on the modularity index or other similarity index to ensure that the nodes within the combined subgraph cluster have high spatio-temporal correlation, so that they can be analyzed and predicted as a whole. The formation of such a cluster not only increases the computational efficiency of the model, but also improves the accuracy of the prediction, because the spatio-temporal patterns within the cluster are more consistent and clear.

[0025] Optionally, in the present embodiment, feature reconstruction is achieved through a time repair strategy, i.e. using a multi-channel sliding window to reconstruct the original data. This process converts the original multi-dimensional state data into a more rich and continuous time series representation, which can better capture the fluctuation trend of charging demand in time and the propagation effect in space. After feature reconstruction, the node features of the spatio-temporal subgraph are more diverse, containing the state changes of the charging pile at different times, which helps the model to understand and predict the spatio-temporal evolution law of charging demand.

[0026] Optionally, in this embodiment, the prediction of charging pile occupancy aims to estimate the probability or proportion of charging piles being occupied by electric vehicles in a certain area within a future period of time (e.g., 3 to 9 steps, i.e., 15 to 45 minutes). The prediction method of the present invention is not limited to instant prediction, but can generate prediction values for multiple future time points through a recursive rolling mechanism. This prediction function is of great significance for the optimal allocation of charging station resources, intelligent scheduling of power grid load, and user charging navigation services, as it provides a forward-looking perspective of charging demand, helping the system and users make better decisions.

[0027] Optionally, in this embodiment, multiple spatio-temporal subgraphs corresponding to multiple charging piles in the area are obtained. First, the algorithm divides the obtained charging pile network data into multiple subgraphs according to the spatio-temporal attributes (such as location, charging demand pattern) of the charging piles through the condensed hierarchical clustering algorithm, each subgraph containing the state data of one or more charging piles, reflecting the spatio-temporal dynamic characteristics of the local network.

[0028] Further, spatio-temporal subgraphs that meet the expected conditions are determined and combined into clusters. The similarity between these spatio-temporal subgraphs will be evaluated, and if the similarity between two subgraphs is higher than a pre-set threshold, they will be identified as meeting the "expected conditions". The subgraphs that meet the conditions will be combined to form one or more spatio-temporal subgraph clusters, which are more conducive to subsequent feature reconstruction and prediction because the spatio-temporal characteristics of the nodes within them are more consistent.

[0029] In addition, feature reconstruction is performed on the spatio-temporal subgraph clusters. Through a time repair strategy, the feature matrix of each subgraph within the cluster is reconstructed. This reconstruction process converts discrete time point data into continuous time series segments, not only increasing the dimension of the input features, but also strengthening the model's ability to model short-term time series dependencies, making it easier to capture real-time changes and long-term trends in charging demand more accurately.

[0030] Furthermore, the spatio-temporal subgraph clusters are used for charging pile occupancy prediction. The spatio-temporal synchronous graph Kolmogorov Arnold Network (STSGKAN) is used as an expert model to perform in-depth analysis on the reconstructed spatio-temporal subgraph clusters and predict the charging pile occupancy at multiple future time points. This prediction process integrates spatial and temporal information, allowing it to consider both the regional characteristics and time fluctuations of charging demand, providing more reliable and detailed prediction results.

[0031] Through the embodiments provided in the present application, the large-scale data of the charging network is "divided and conquered, fused and cooperated". Through hierarchical clustering, the charging pile network is decomposed into multiple spatiotemporal subgraphs that can be independently analyzed. These subgraphs are then integrated into spatiotemporal subgraph clusters according to their similarity to achieve efficient allocation of computing resources. Feature reconstruction technology enhances the expression of information in the subgraph clusters, enabling a more comprehensive reflection of the spatiotemporal characteristics of charging demand. Finally, using the reconstructed spatiotemporal subgraph clusters for prediction not only overcomes the limitations of traditional prediction methods in terms of computational efficiency and spatiotemporal feature fusion, but also significantly improves the accuracy and timeliness of the prediction, especially when dealing with complex urban-level charging network data. This method provides key technical support for the operation and management of charging infrastructure and user services, helping to achieve intelligent scheduling of the power grid and efficient use of charging resources, while also providing reliable prediction information for users planning charging routes.

[0032] As an optional solution, before combining at least two spatiotemporal subgraphs to obtain a spatiotemporal subgraph cluster, the method further comprises:

[0033] determining a central spatiotemporal subgraph corresponding to a central charging pile from the plurality of spatiotemporal subgraphs;

[0034] determining candidate spatiotemporal subgraphs with a similarity greater than a preset threshold from the plurality of spatiotemporal subgraphs;

[0035] In the case where the number of candidate spatiotemporal subgraphs is greater than or equal to a subgraph threshold, the central spatiotemporal subgraph and the subgraph threshold number of spatiotemporal subgraphs with the highest similarity in the candidate spatiotemporal subgraphs are determined as the spatiotemporal subgraph cluster.

[0036] Optionally, in the present embodiment, a "central spatiotemporal subgraph" is selected from the plurality of spatiotemporal subgraphs. This subgraph usually contains the central charging pile and other charging piles with strong spatiotemporal relevance. The selection of the central subgraph is based on its location importance, representative of charging demand, or pivotal role in the network.

[0037] Next, the algorithm identifies those subgraphs from the remaining plurality of spatiotemporal subgraphs that have a similarity greater than a preset threshold to the central spatiotemporal subgraph, marking them as "candidate spatiotemporal subgraphs". This step ensures that the subgraphs within the cluster all have similar temporal fluctuation characteristics and spatial relevance, which helps improve the accuracy of subsequent feature fusion and prediction.

[0038] If the number of candidate spatiotemporal subgraphs reaches or exceeds the set subgraph threshold, the algorithm selects the subgraph threshold number of subgraphs with the highest similarity from these candidate subgraphs, together with the central spatiotemporal subgraph Figure 1 , to construct the final "spatiotemporal subgraph cluster". The construction of the cluster considers the optimal combination of subgraphs, aiming to form a subgraph set that contains sufficient information and is easy to process efficiently.

[0039] It can be understood that constructing the spatio-temporal subgraph cluster is a key preprocessing step, which directly affects the efficiency and accuracy of model prediction. By carefully selecting a "central spatio-temporal subgraph", the algorithm can focus on the most active or important local area in the charging network. Then, the process of screening "candidate spatio-temporal subgraphs" further refines the selection criteria of subgraphs, ensuring that the spatio-temporal features of subgraphs within the cluster are highly consistent, so that the regional commonality and individual differences of charging demand can be more effectively captured in the subsequent feature reconstruction and prediction.

[0040] In the process of constructing the spatio-temporal subgraph cluster, the setting of the subgraph threshold plays a role in balancing the number of subgraphs and the quality of the cluster. Too many subgraphs will lead to excessive consumption of computing resources, while too few subgraphs may not fully reflect the regional characteristics. Reasonable setting of the subgraph threshold ensures that the cluster not only covers the necessary information dimensions, but also maintains appropriate computational complexity. The most similar subgraphs are selected into the cluster, which not only simplifies the information fusion between subgraphs, but also improves the ability of the prediction model to capture the spatio-temporal evolution law of charging demand.

[0041] Through the selection of the central spatio-temporal subgraph and the clustering of the candidate subgraph, the embodiments provided by the present application achieve precise focus on key areas in large-scale charging networks. The formed spatio-temporal subgraph cluster is not only rich in information but also easy to process, providing an optimized foundation for subsequent feature reconstruction and charging pile occupancy rate prediction, effectively solving the contradiction between computational efficiency and prediction accuracy, and significantly improving the practical value of the prediction model in the analysis of urban-level charging networks.

[0042] As an optional solution, after determining the candidate spatio-temporal subgraphs with a similarity greater than the preset threshold from the plurality of spatio-temporal subgraphs, the method further includes:

[0043] In the case where the number of candidate spatio-temporal subgraphs is less than the subgraph threshold, the spatio-temporal subgraph with the greatest similarity in the candidate spatio-temporal subgraphs is determined as a new central spatio-temporal subgraph, and new candidate spatio-temporal subgraphs with a similarity greater than the preset threshold to the new central spatio-temporal subgraph are determined from the plurality of spatio-temporal subgraphs;

[0044] At least one spatio-temporal subgraph from the central spatio-temporal subgraph, the candidate spatio-temporal subgraph, and the new candidate spatio-temporal subgraph is determined as the spatio-temporal subgraph cluster.

[0045] Optionally, in this embodiment, candidate spatio-temporal subgraphs with a similarity greater than the preset threshold to the central spatio-temporal subgraph are screened from the plurality of spatio-temporal subgraphs. If the number of these screened candidate subgraphs meets or exceeds the standard set by the subgraph threshold, they are directly combined with the central spatio-temporal subgraph to form the spatio-temporal subgraph cluster, entering the feature reconstruction and prediction phase.

[0046] When the number of candidate spatiotemporal subgraphs is less than the subgraph threshold, it indicates that the current cluster is not large enough to form an effective prediction model, and further expansion of the cluster is needed. At this time, the system will take the following steps:

[0047] The spatiotemporal subgraph with the highest similarity among the candidate spatiotemporal subgraphs is re-designated as the center spatiotemporal subgraph. The purpose of this is to try to use a subgraph with higher similarity to more subgraphs as a new center to enhance the representativeness and prediction ability of the cluster.

[0048] Around this new center, the system will again review all spatiotemporal subgraphs to find subgraphs with a similarity greater than the preset threshold to the new center subgraph as "new candidate spatiotemporal subgraphs".

[0049] Finally, the new center spatiotemporal subgraph, the original candidate spatiotemporal subgraph, and at least one of the newly found candidate spatiotemporal subgraphs are integrated together to determine the final spatiotemporal subgraph cluster. In this way, even if the cluster size is insufficient in the initial screening, by dynamically adjusting the center subgraph, the system can build a cluster containing sufficient similar subgraphs to support subsequent prediction.

[0050] It should be noted that the embodiment adopts a flexible dynamic adjustment mechanism to cope with the fluctuation of the number of subgraphs in different charging network environments. First, an initial cluster is tried to be formed through the screening and combination of "center spatiotemporal subgraphs" and "candidate spatiotemporal subgraphs". However, when the number of candidate subgraphs is insufficient, the system will dynamically reselect the center subgraph, usually selecting the spatiotemporal subgraph with the highest similarity among all candidates, and then perform the screening process again around this new center until the cluster built meets the requirements of the subgraph threshold.

[0051] Through the embodiments provided by the present application, the dynamic adjustment mechanism ensures that the spatiotemporal subgraph cluster can not only reflect the typical characteristics of regional charging demand, but also cover a sufficient number of subgraphs to achieve the stability and accuracy of prediction. Most importantly, it provides an adaptive solution for the formation of a spatiotemporal subgraph cluster, which can quickly adjust and form an effective prediction model cluster even in the case of complex and variable charging network topology. This method not only improves the adaptability and generalization ability of the prediction model, but also provides a solid foundation for the charging demand prediction of large-scale charging networks, which helps to realize the intelligent management and optimization of charging infrastructure and the refinement of user navigation services.

[0052] As an optional solution, the feature reconstruction of each spatiotemporal subgraph in the spatiotemporal subgraph cluster includes:

[0053] The original features of each spatio-temporal subgraph are slidingly reorganized by a multi-channel sliding window with a preset step size, to obtain a spatio-temporal subgraph after feature reorganization, wherein the number of feature nodes corresponding to the spatio-temporal subgraph after feature reorganization is N times the number of feature nodes of the spatio-temporal subgraph before feature reorganization, and N is the number of channels of the multi-channel sliding window.

[0054] Optionally, in the embodiment, before feature reconstruction, the charging pile network has been divided into a plurality of spatio-temporal subgraphs by a condensed hierarchical clustering algorithm. Each subgraph represents a group of charging piles with strong spatio-temporal correlation, and the state data thereof will be used for subsequent analysis and prediction.

[0055] For each subgraph in the spatio-temporal subgraph cluster, a multi-channel sliding window technique with a preset step size is used for sliding reorganization of features. This process involves sliding the window in the time dimension to cover consecutive time points and organizing and integrating the data at these time points by channel.

[0056] Through the sliding reorganization of the multi-channel sliding window, the number of original feature nodes of each spatio-temporal subgraph is expanded by N times, where N is the number of channels of the sliding window. This operation increases the feature dimension processed by the model, enabling the model to consider information at multiple time points simultaneously and improving the modeling capability for short-term time series dependence.

[0057] Each spatio-temporal subgraph after reorganization contains N times the number of feature nodes of the original subgraph, that is, each subgraph now contains richer time series features. These spatio-temporal subgraphs after feature reorganization will serve as input for model training, enabling more accurate reflection of the spatio-temporal variation of charging pile occupancy.

[0058] Through the embodiments provided by the present application, by applying a multi-channel sliding window technique with a preset step size to each subgraph in the spatio-temporal subgraph cluster, the original features can be effectively reorganized, converting the state data of charging piles in each subgraph from individual time point information to multi-dimensional representation of consecutive time segments. This expansion of feature node number (N times the original number, N is the number of channels) not only enhances the model's ability to capture short-term fluctuations in charging demand, but also improves the understanding of long-term trends through the introduction of time channels, which is crucial for predicting the spatio-temporal evolution of electric vehicle charging demand. The spatio-temporal subgraph after feature reorganization has a more comprehensive time series representation, forming a fusion of frequency and time domain features, which enables the prediction model to more accurately predict the occupancy rate of charging piles at multiple future time steps, providing a scientific basis for the rational planning and operation of charging infrastructure.

[0059] As an optional solution, using the reconstructed spatio-temporal subgraph cluster, the occupancy rate of charging piles in a region is predicted, including:

[0060] According to each reconstructed spatio-temporal subgraph in the reconstructed spatio-temporal subgraph cluster, a plurality of prediction sub-models are trained for the plurality of charging piles, wherein one prediction sub-model corresponds to one reconstructed spatio-temporal subgraph;

[0061] The occupancy rates of the plurality of charging piles in the region are predicted using the plurality of prediction sub-models.

[0062] Optionally, in this embodiment, after the feature reconstruction is completed, for each reconstructed spatio-temporal subgraph in the spatio-temporal subgraph cluster, a dedicated prediction sub-model is trained using the rich features contained therein. This means that we will have a group of expert models, each model focusing on its corresponding subgraph. This "divide and conquer" approach reduces computational complexity while ensuring that each sub-model can accurately capture the spatio-temporal dynamics of its subgraph.

[0063] With the prediction sub-models for each subgraph, the next step is to use these sub-models to predict the occupancy rates of the plurality of charging piles in the region. The prediction process is based on the deep understanding of the features contained in the reconstructed spatio-temporal subgraph and the learning of charging behavior patterns.

[0064] Since each prediction sub-model only focuses on one subgraph, we also need a mechanism to integrate the prediction results of these models to obtain the charging pile occupancy rate prediction for the entire region. This mechanism dynamically integrates the outputs of each expert model through a gating network and optimizes the global output using a transfer learning mechanism to ensure that the prediction results are accurate and comprehensive, effectively capturing the global and local spatio-temporal evolution rules of charging demand.

[0065] Through the embodiments provided in this application, by customizing a prediction sub-model for each subgraph, we can not only take advantage of the unique spatio-temporal characteristics of each subgraph for fine-grained prediction, but also avoid the computational bottleneck of global models when facing large-scale networks. The prediction results of this series of sub-models are then integrated through a dynamic integration mechanism to ensure the global consistency of the final prediction while preserving the local characteristics of each subgraph. This method not only guarantees prediction accuracy but also significantly reduces computational complexity, providing a feasible solution for real-time prediction of large-scale charging networks. Through this approach, we can effectively capture the spatio-temporal fluctuations of charging demand and provide accurate data support for the optimization of charging infrastructure and user navigation services.

[0066] As an optional solution, using a plurality of prediction sub-models to predict the occupancy rates of a plurality of charging piles in a region includes:

[0067] In the process of using a plurality of prediction sub-models to predict the occupancy rate of a target charging pile in a region, a plurality of prediction results are obtained by using the plurality of prediction sub-models to predict the occupancy rate of the target charging pile, wherein the plurality of charging piles include the target charging pile;

[0068] The multiple prediction results are weighted and summed to obtain the prediction results of the multiple prediction sub-models on the occupancy rate of the target charging pile.

[0069] Optionally, in the embodiment, after the feature reconstruction and model training stage, each prediction sub-model can independently predict the occupancy rate of the target charging pile in the spatio-temporal sub-graph cluster. Each sub-model gives a prediction result based on its training data and algorithm design, and these results reflect the predicted patterns and trends of charging demand in different sub-graphs.

[0070] After each sub-model completes the prediction, the prediction results of the target charging pile occupancy rate are collected. These results may show certain differences, because each sub-model focuses on different spatio-temporal sub-graphs, and the charging pile demand characteristics in these sub-graphs may be different.

[0071] The weights of the prediction results of each prediction sub-model are dynamically determined using a gating network, and then these weighted prediction results are summed to obtain the final prediction result of the occupancy rate of the target charging pile. The gating network dynamically adjusts the weights by learning the importance of different sub-graphs in prediction, ensuring that the prediction result not only considers the prediction information of multiple models, but also optimizes according to the spatio-temporal characteristics of the sub-graphs, improving the accuracy and reliability of the prediction.

[0072] Through the embodiments provided in the present application, each prediction sub-model can independently predict the occupancy rate of the target charging pile by deeply analyzing the reconstructed spatio-temporal sub-graphs. However, due to the differences in characteristics of different sub-graphs, the prediction result of a single sub-model may be limited by its local perspective. Therefore, by dynamically weighting and summing the outputs of multiple prediction sub-models through a gating network, the prediction information of all expert models can be considered comprehensively, and the weights can be adjusted according to the spatio-temporal correlation of the sub-graphs, ensuring that the final prediction result is both comprehensive and accurate. This method overcomes the limitations of traditional prediction models in handling large-scale charging networks, improves the ability to capture the spatio-temporal evolution law of charging demand, and provides a solid data foundation for efficient scheduling of charging facilities and optimization of user charging behavior. Through the dynamic fusion mechanism, we not only can utilize the advantages of each sub-model, but also can effectively avoid the overall prediction distortion caused by the prediction bias of a single model, thereby achieving a double improvement in prediction accuracy and efficiency.

[0073] As an optional solution, the method further comprises:

[0074] According to a preset period, a first number of reconstructed spatio-temporal sub-graph clusters are obtained;

[0075] The occupancy rates of the plurality of charging piles in the region are predicted according to each reconstructed spatio-temporal subgraph in the first number of reconstructed spatio-temporal subgraph clusters using a plurality of prediction sub-models, to obtain occupancy rate prediction results of the plurality of charging piles, wherein the occupancy rate prediction result of one charging pile includes second number of occupancy rate prediction sub-results, and the second number is less than the first number.

[0076] Optionally, in the present embodiment, a plurality of reconstructed spatio-temporal subgraph clusters are collected according to a preset period, and each cluster represents a snapshot of the spatio-temporal characteristics of the charging pile network and the charging demand pattern in the period. The purpose of this is to analyze and predict the change law of charging demand over time and the occupancy of charging piles in different time periods.

[0077] For each collected reconstructed spatio-temporal subgraph cluster, an occupancy rate prediction is performed using a previously trained prediction sub-model. Here, the prediction sub-model, based on its learning of historical charging pile data and understanding of the spatio-temporal characteristics of charging behavior, can give a prediction result of the occupancy rate of the charging pile in each subgraph.

[0078] The prediction results of the first number of reconstructed spatio-temporal subgraph clusters by the prediction sub-models are integrated to generate occupancy rate prediction results containing a plurality of prediction periods. It is worth noting that the occupancy rate prediction result of each charging pile includes second number of occupancy rate prediction sub-results, which means that even when processing clusters of multiple time periods, the prediction of each charging pile is refined to a shorter time step, ensuring the detail and timeliness of the prediction results.

[0079] Through the embodiments provided by the present application, by periodically obtaining a plurality of reconstructed spatio-temporal subgraph clusters, we can establish snapshots of the spatio-temporal characteristics of the charging pile network in different time periods, thereby gaining a deep understanding of the law of charging demand change over time. Using the trained prediction sub-models to predict the occupancy rates of these clusters can generate occupancy rate prediction results covering multiple prediction periods. In particular, for each charging pile, its occupancy rate prediction result is refined to include prediction sub-results in a second number of shorter periods, which not only reflects the strong flexibility of the prediction model, but also ensures the comprehensiveness and accuracy of the charging demand prediction, providing strong support for the dynamic management of charging infrastructure and the planning of user charging behavior. This method combines the stability of long periods with the sensitivity of short periods through periodic data acquisition and prediction, achieving a comprehensive grasp of the spatio-temporal characteristics of charging demand, and creating conditions for the efficient operation of the electric vehicle charging network.

[0080] As an optional solution, the above charging pile occupancy rate prediction method is applied to the time and space synchronous prediction based on hierarchical clustering and expert hybrid. In this scenario, to solve the problems of low calculation efficiency of traditional graph neural network and difficulty in time and space feature fusion in large-scale charging pile network, the embodiment divides the global network into multiple time and space related subgraphs through the agglomerative hierarchical clustering algorithm, effectively capturing the dynamic characteristics of local areas; adopts a time repair strategy to reconstruct the feature matrix, enhancing the modeling ability of short-term time series dependence; innovatively designs a time and space synchronous graph Kolmogorov-Arnold network as an expert model to replace the traditional spline function to realize frequency domain feature fusion, significantly improving the time and space feature extraction efficiency; through a gating network, the prediction results of each subgraph expert model are dynamically integrated, and a transfer learning mechanism is used to optimize the global output. This method significantly reduces the calculation complexity while ensuring the prediction accuracy, can accurately capture the time and space evolution law of charging demand, and is suitable for scenarios such as power grid load scheduling, charging station resource optimization, and user navigation services, providing efficient decision support for large-scale electric vehicle charging infrastructure.

[0081] It should be noted that in the global energy transformation wave, electric vehicles have become a key carrier for decarbonization in the transportation sector due to their zero-emission characteristics. With the introduction of national schedules for banning the sale of gasoline vehicles, the market penetration rate of electric vehicles continues to rise, and the scale of charging infrastructure is expanding exponentially. This expansion is not only reflected in the rapid increase in the number of charging piles, but also in the qualitative change in the time and space complexity of charging behavior: user charging demand has shifted from fixed locations to global roaming, and charging behavior is deeply coupled with urban commuting tides, business activity cycles, and even weather events. Traditional prediction models based on single-point statistics have shown fundamental limitations in this context - they cannot capture the networked transmission effects of charging demand in space and the nonlinear fluctuation characteristics in time.

[0082] The essence of charging demand prediction is to deconstruct the dynamic interaction of the "person-vehicle-pile-grid" multi-element system. When the scale of charging pile nodes exceeds ten thousand, the spatial correlation between nodes no longer follows a simple distance decay law, but rather a complex network topology with multiple centers and multiple levels. At the same time, the randomness of user behavior causes charging demand to have both short-term burstiness and long-term regularity in the time dimension. This high-dimensional time and space coupling characteristic significantly increases the prediction bias of traditional time series models and machine learning methods, especially during peak hours, where prediction errors can directly cause local power grid overload or charging resource idleness.

[0083] Current mainstream prediction methods can be categorized into three types, each with insurmountable bottlenecks:

[0084] The first type is the simulation method based on physical modeling. This method tries to deduce demand changes by establishing physical equations of charging behavior, such as modeling the charging process as a Poisson distribution event stream. Although it has theoretical self-consistency, its core defect is the oversimplification of real-world scenarios: on the one hand, user decisions are influenced by multiple factors such as price fluctuations, queue length, travel planning, etc., making it difficult to characterize with fixed parameter equations; on the other hand, the dynamic nature of urban road networks results in non-stationary characteristics of spatial transmission of charging demand, and static physical models cannot capture this dynamic correlation. More importantly, when the node scale expands, the computational complexity of equation solving increases exponentially, making real-time prediction an impossible task.

[0085] The second type is the data-driven method based on traditional machine learning. Algorithms represented by support vector machines (SVM) and random forests train prediction models through historical data, to some extent overcoming the rigid defects of physical models. However, this method has two major drawbacks: first, feature engineering dependence, the model's effectiveness is highly dependent on manually designed spatiotemporal features (such as regional charging peak periods, adjacent pile group correlation, etc.), and the feature dimension in a city-level network can be millions, making manual design neither comprehensive nor sustainable; second, spatiotemporal feature fragmentation, most models use a serial architecture of "spatial clustering first, then temporal prediction", which results in the interaction effect of spatial dependence and temporal dynamics being ignored. For example, during the morning rush hour, the surge in demand for charging in business districts will be transmitted through the road network to adjacent residential areas, and this spatial spillover effect overlaps with the time peak, which cannot be captured by serial models.

[0086] The third type is a deep learning method based on graph neural network. As the current frontier technology, graph convolution network (GCN) and graph attention network (GAT) model the node space relationship through graph structure, which significantly improves the spatial feature extraction capability. However, in-depth analysis shows that the existing graph neural network faces three dilemmas: computational efficiency dilemma: graph convolution operation needs to rely on adjacency matrix operation, when the number of nodes reaches ten thousand, the memory and computing power consumed by matrix storage and calculation far exceed the bearing capacity of ordinary servers. The test data shows that the training time of one hundred thousand nodes can reach dozens of hours, which cannot meet the demand of power grid response in minutes. Time and space asynchronous dilemma: although the mainstream architecture such as STGCN tries to integrate time and space features, it still adopts the cascade design of “spatial convolution layer + time convolution layer”. This architecture causes the time information to be frozen when extracting spatial features, and the spatial relationship to be solidified when extracting time features, which cannot model the collaborative evolution law of charging demand in the time-space continuum. The typical performance is that the prediction of “tidal effect” is significantly biased--the demand peak of early peak from residential area to business district is often predicted as double peak or delayed peak. Frequency domain modeling dilemma: the latest research tries to realize time and space synchronous modeling through spline function, but the spline function has the problem of loss of high frequency components in frequency domain feature extraction, and the calculation of its piecewise polynomial leads to low efficiency of back propagation. This makes it difficult for the model to capture the sudden fluctuations in charging demand, while significantly slowing down the training speed.

[0087] In order to solve these systematic bottlenecks, the embodiment realizes computational relief by dividing time-space subgraphs through hierarchical clustering, integrates frequency domain features by using Fourier domain time-space synchronous network (STSGKAN), and dynamically aggregates local prediction results based on the gating weighted expert integration mechanism to form an integrated prediction architecture of “divide and conquer-fusion-collaboration”, realizing revolutionary leap in computational efficiency and essential breakthrough in prediction accuracy.

[0088] The embodiment proposes a method and system (MOE-STSGKAN) for electric vehicle charging demand prediction based on hierarchical clustering and expert hybridization. The core of the method is to solve the inherent contradiction between computational efficiency and time-space feature synchronization in large-scale charging network through an integrated architecture of “divide and conquer-fusion-collaboration”. The method first combines agglomerative hierarchical clustering algorithm and time patching to reconstruct historical input features into multiple local time-space subgraphs; then, it uses STSGKAN for pre-training, saves and defines each model as an expert model. Finally, it trains the historical input features by combining all fixed parameter expert models and fine-tuning gating network to obtain the final prediction value. The method realizes a revolutionary breakthrough in computational efficiency, achieves an essential leap in prediction accuracy, and gives the system the generalization advantage of dynamic expansion, completely solving the three problems of efficiency, accuracy and generalization in large-scale charging network prediction. The method includes the following steps:

[0089] The charging pile network is divided into multiple spatiotemporal correlation subgraphs by using a condensed hierarchical clustering algorithm; each subgraph is reconstructed in features by using a time repair strategy; a spatiotemporal synchronization graph Kolmogorov-Arnold network (STSGKAN) expert model is constructed for each subgraph; the outputs of the expert models are dynamically fused through a gating network; and the charging pile occupancy rate in multiple future time steps is predicted.

[0090] Specifically, the embodiment discloses a spatiotemporal synchronization prediction method based on hierarchical clustering and expert mixing, and a flowchart is as shown in Figure 2 The flow steps include:

[0091] S1, the charging pile network is divided into multiple spatiotemporal correlation subgraphs by using a condensed hierarchical clustering algorithm;

[0092] S2, each subgraph is reconstructed in features by using a time repair strategy;

[0093] S3, a spatiotemporal synchronization STSGKAN expert model is constructed for each subgraph;

[0094] S4, the outputs of the expert models are dynamically fused through a gating network;

[0095] S5, the charging pile occupancy rate in multiple future time steps is predicted.

[0096] The specific implementation process is as follows:

[0097] I. System architecture and hardware deployment

[0098] The embodiment is preferably deployed on a high-performance computing platform equipped with an NVIDIA GeForce RTX 4090 GPU, and distributed training is realized by using a PyTorch 2.0+ framework. The system adopts a modular design and includes a data acquisition module, a preprocessing engine, a spatiotemporal graph constructor, a hierarchical clustering processor, a feature reconstruction unit, an STSGKAN expert model cluster, a gating network fusioner, and a prediction output interface. The data acquisition module acquires multi-dimensional state data of charging piles in real time through a city charging management platform API, with a sampling frequency of 5 minutes / time, covering key fields such as charging power, occupancy state, geographic location coordinates, cumulative charging duration, and surrounding traffic congestion index. The preprocessing engine uses a spatiotemporal weighted mean imputation method to process missing values, and dynamically adjusts the spatiotemporal weight through an exponential decay function:

[0099]

[0100] wherein represents a time interval, represents a spatial distance, minutes, and km respectively control the time and spatial decay rate. Normalization linearly maps the charging demand value to the interval [0, 1] to eliminate the dimensional difference interference on model training. The space-time graph constructor abstracts the urban traffic network as an undirected graph , the node set contains traffic analysis areas, and the edge set defines the region connectivity relationship. The adjacency matrix A adopts thresholding processing: if two regions have a direct connection through a trunk road and the distance is less than 3 km, then , if connected through a secondary road, then , otherwise 0. The feature matrix integrates 12-dimensional time series features, including the charging pile occupancy rate change trend, power fluctuation characteristics and traffic flow dynamic parameters within 1 hour.

[0101] II. Hierarchical clustering subgraph division

[0102] The global charging network is divided into 4 subgraphs with strong space-time correlation by the agglomerative hierarchical clustering algorithm. The algorithm adopts a bottom-up merging strategy. In the initialization stage, each charging area is regarded as an independent subgraph, and the similarity between subgraphs is calculated based on the maximum modularity criterion:

[0103]

[0104] wherein is the adjacency matrix of the network, and are the degrees of nodes and , is the total number of edges in the network, is an indicator function, and if and belong to the same subgraph, then 1, otherwise 0. The iterative process continues to merge the subgraph pair with the highest similarity until the preset subgraph number threshold is reached. The final subgraph cluster includes this partitioning strategy not only reduces the computational complexity to , but also ensures that the nodes within the subgraph have strong space-time correlation, laying a foundation for subsequent expert model training.

[0105] III. Reconstruction of space-time features

[0106] The time repair strategy reconstructs the feature matrix through a multi-channel sliding window to solve the problem of time-space feature fragmentation in traditional methods. For the original feature of each subgraph , a 3-channel sliding window with a step size of is used to reorganize it into:

[0107]

[0108] Since the original input feature node number is , the reconstructed new input feature node number is . After reconstruction, the new adjacency matrix represents the connection mode of each channel graph structure as shown in Figure 3 . This operation converts discrete time points into continuous time segments, enabling the model to capture the phase continuity of charging demand evolution.

[0109] Four, STSGKAN expert model construction

[0110] After obtaining multiple subgraph features, a spatio-temporal synchronous graph Kolmogorov-Arnold network (STSGKAN) is used to extract deep features from each subgraph and predict the future charging pile demand power of each node in each subgraph.

[0111] STSGKAN is a graph convolutional network based on KAN, which accepts node features and adjacency matrices as input and transmits information through multiple KAN layers, and finally obtains output through a linear layer.

[0112] Assuming the input is , the adjacency matrix is , and there are KAN layers, then the forward propagation of the entire STSGKAN can be represented as:

[0113]

[0114] In the formula, is the hidden state of the layer, , and are the weight matrices of the input layer, the intermediate layer and the output layer, is the bias of the layer, is the final output, is the KAN internal nonlinear activation function, which uses sine and cosine functions to interact with input features, and combines the results of these functions through predefined Fourier coefficients. The process will be described more accurately below.

[0115] In the KAN layer, the input data is converted into a higher-dimensional space, where each dimension corresponds to a different frequency component. This process can be approximated by a Fourier series. First, a parameter is set to store the Fourier coefficients, which has a dimension of . The first dimension (size 2) corresponds to the coefficients of the cosine and sine terms, the second dimension (size ) corresponds to the dimension of the output feature, and the third dimension (size ) corresponding to the dimensions of the input features, the last dimension (size ) represents the number of frequency components; then for each input feature generate a sequence of integers from 1 to ; finally, multiply and sum the Fourier coefficients with these sine and cosine values:

[0116]

[0117] where and are the coefficients extracted from and , respectively. In this way, KAN can convert input features into representations in the frequency domain, thus more effectively capturing periodic and trend features in the data. This frequency domain fusion method not only improves the feature expression ability of the model, but also reduces the training difficulty, so that the model can converge faster.

[0118] After establishing the STSGKAN prediction model corresponding to the subgraph through the above process, the STSGKAN prediction model is trained with MSE as the loss function and Adam as the parameter update optimization algorithm. Then save the trained parameters. Each trained model will be used as an expert model for subsequent transfer learning.

[0119] Five, gating network fusion mechanism

[0120] The gating network realizes the dynamic weighted fusion of the output of the expert model, and its structure contains two layers of fully connected networks. The first layer compresses the feature dimension through the dimension reduction matrix, and the second layer generates the expert weight:

[0121]

[0122] where is the output sequence of the gating network, and are the weights of the two fully connected layers, and the final prediction is realized through the spatiotemporal adaptive weighting.

[0123] Six, multi-step prediction and engineering implementation

[0124] The prediction module realizes multi-step output by using the recursive rolling mechanism: input 1-hour historical data (12 time steps), add it to the input sequence after sliding window update, and iteratively generate 3-9 step (15-45 minutes) predictions. The output post-processing includes denormalization ​Y = X norm(X max-X min) + X min and physical constraint correction, to ensure that the predicted value meets the actual physical boundary. The engineering deployment stage realizes operator fusion and inter-layer optimization through the TensorRT engine, and converts the calculation graph into a highly optimized inference engine.

[0125] Through the embodiments provided in the present application, for the calculation efficiency bottleneck, the present embodiment innovates the condensed hierarchical clustering-temporal-spatial subgraph partitioning technology. This technology breaks through the global calculation paradigm of traditional graph neural networks, and decomposes the large-scale charging network into multiple highly cohesive subgraphs through the modularity optimization criterion. The nodes inside each subgraph have strong temporal-spatial correlation, and the subgraphs are moderately decoupled. This "divide and conquer" strategy reduces the computational complexity from quadratic to linear logarithmic, laying the foundation for real-time prediction. To solve the problem of temporal-spatial feature fragmentation, the present embodiment develops a Fourier domain space-time synchronous graph network (STSGKAN). This architecture realizes true space-time coupling modeling through three innovations: reconstructing the time series features with a multi-channel sliding window, converting discrete time points into continuous time segments, and capturing the phase continuity of demand evolution; replacing the traditional spline function with a Fourier basis function to realize lossless extraction of spectral features through a sine-cosine orthogonal basis; constructing a learnable Fourier coefficient tensor to dynamically adjust the weight proportion of different frequency components in the space-time features. To balance global consistency and local characteristics, the present embodiment designs a gated weighted expert integration mechanism: each space-time subgraph trains a dedicated STSGKAN expert model to deeply mine local space-time patterns; the gating network dynamically integrates expert outputs through an attention mechanism to generate a globally optimal prediction; and a transfer learning strategy is used to freeze the expert parameters and only fine-tune the gating network to achieve efficient knowledge transfer.

[0126] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown.

[0127] The present application also provides a charging pile occupancy rate prediction device. It should be noted that the charging pile occupancy rate prediction device of the present application can be used to execute the charging pile occupancy rate prediction method provided by the present application. The charging pile occupancy rate prediction device provided by the present application is introduced as follows.

[0128] Figure 4 is a schematic diagram of the charging pile occupancy rate prediction device according to the present application. As shown in Figure 4 , the device comprises:

[0129] The acquisition unit 401 is configured to acquire a plurality of spatio-temporal subgraphs corresponding to a plurality of charging piles in a region, wherein one spatio-temporal subgraph is used to indicate at least two items of state data of one charging pile.

[0130] The combination unit 402 is configured to, in a case where at least two spatio-temporal subgraphs meeting an expected condition are determined from the plurality of spatio-temporal subgraphs, combine the at least two spatio-temporal subgraphs to obtain a spatio-temporal subgraph cluster, wherein the expected condition is used to indicate that a similarity degree of the at least two spatio-temporal subgraphs is greater than a preset threshold.

[0131] The reconstruction unit 403 is configured to perform feature reconstruction on each spatio-temporal subgraph in the spatio-temporal subgraph cluster, wherein the reconstructed spatio-temporal subgraph indicates more items of state data of the charging pile.

[0132] The prediction unit 404 is configured to perform occupancy rate prediction of the charging pile in the region by using the reconstructed spatio-temporal subgraph cluster.

[0133] As an optional solution, the apparatus further includes:

[0134] The first determination module is configured to, before the at least two spatio-temporal subgraphs are combined to obtain the spatio-temporal subgraph cluster, determine a center spatio-temporal subgraph corresponding to a center charging pile from the plurality of spatio-temporal subgraphs.

[0135] The second determination module is configured to, before the at least two spatio-temporal subgraphs are combined to obtain the spatio-temporal subgraph cluster, determine a candidate spatio-temporal subgraph having a similarity degree greater than a preset threshold from the plurality of spatio-temporal subgraphs.

[0136] The third determination module is configured to, before the at least two spatio-temporal subgraphs are combined to obtain the spatio-temporal subgraph cluster, in a case where a quantity of the candidate spatio-temporal subgraphs is greater than or equal to a subgraph threshold, determine a subgraph threshold quantity of spatio-temporal subgraphs having the highest similarity degree from the center spatio-temporal subgraph and the candidate spatio-temporal subgraph as the spatio-temporal subgraph cluster.

[0137] As an optional solution, the apparatus further includes:

[0138] The fourth determination module is configured to, after the candidate spatio-temporal subgraph having the similarity degree greater than the preset threshold is determined from the plurality of spatio-temporal subgraphs, in a case where the quantity of the candidate spatio-temporal subgraphs is less than the subgraph threshold, determine a spatio-temporal subgraph having the greatest similarity degree from the candidate spatio-temporal subgraphs as a new center spatio-temporal subgraph, and determine a new candidate spatio-temporal subgraph having a similarity degree greater than the preset threshold from the plurality of spatio-temporal subgraphs in relation to the new center spatio-temporal subgraph.

[0139] The fifth determination module is configured to, after the candidate spatio-temporal subgraph having the similarity degree greater than the preset threshold is determined from the plurality of spatio-temporal subgraphs, determine at least one spatio-temporal subgraph from the center spatio-temporal subgraph, the candidate spatio-temporal subgraph, and the new candidate spatio-temporal subgraph as the spatio-temporal subgraph cluster.

[0140] As an optional solution, the reconstruction unit 403 comprises:

[0141] a reconstruction module, configured to perform sliding recombination on the original features of each spatio-temporal subgraph according to a preset step size and a multi-channel sliding window, to obtain a spatio-temporal subgraph after feature recombination, wherein the number of feature nodes corresponding to the spatio-temporal subgraph after feature recombination is N times the number of feature nodes of the spatio-temporal subgraph before feature recombination, and N is the number of channels of the multi-channel sliding window.

[0142] As an optional solution, the prediction unit 404 comprises:

[0143] a training module, configured to train a plurality of prediction sub-models for the plurality of charging piles according to each reconstructed spatio-temporal subgraph in the reconstructed spatio-temporal subgraph cluster, wherein one prediction sub-model corresponds to one reconstructed spatio-temporal subgraph;

[0144] a prediction module, configured to use the plurality of prediction sub-models to predict the occupancy rates of the plurality of charging piles in the region.

[0145] As an optional solution, the prediction module comprises:

[0146] a first obtaining sub-module, configured to, in the process of predicting the occupancy rate of a target charging pile in the region using the plurality of prediction sub-models, obtain a plurality of prediction results obtained by the plurality of prediction sub-models in predicting the occupancy rate of the target charging pile, wherein the plurality of charging piles comprises the target charging pile;

[0147] a fusion sub-module, configured to perform weighted summation on the plurality of prediction results, to obtain an occupancy rate prediction result of the target charging pile by the plurality of prediction sub-models.

[0148] As an optional solution, the device further comprises:

[0149] a second obtaining sub-module, configured to obtain a first number of reconstructed spatio-temporal subgraph clusters according to a preset period;

[0150] a prediction sub-module, configured to use the plurality of prediction sub-models to predict the occupancy rates of the plurality of charging piles in the region according to each reconstructed spatio-temporal subgraph in the first number of reconstructed spatio-temporal subgraph clusters, to obtain occupancy rate prediction results of the plurality of charging piles, wherein the occupancy rate prediction result of one charging pile comprises a second number of occupancy rate prediction sub-results, and the second number is less than the first number.

[0151] The charging pile occupancy rate prediction device comprises a processor and a memory, and the above-mentioned obtaining unit, extracting unit, recursive unit, determining unit, etc. are all stored in the memory as program units, and the corresponding functions are realized by the processor executing the above-mentioned program units stored in the memory.

[0152] The processor comprises a core, and the core retrieves corresponding program units in the memory.

[0153] The memory can comprise a non-permanent memory in a computer readable medium, a random access memory (RAM) and / or a non-volatile memory such as a read-only memory (ROM) or a flash memory (flash RAM), and the memory comprises at least one memory chip.

[0154] The embodiment of the present application provides a computer readable storage medium, which stores a program, and the program is executed by a processor to realize the charging pile occupancy prediction method.

[0155] The embodiment of the present application provides a processor, which is used for running a program, wherein the program is executed to realize the charging pile occupancy prediction method.

[0156] As shown in the figure, the embodiment of the present application provides an electronic device, which comprises a processor, a memory and a program stored in the memory and capable of running on the processor, and the processor realizes the following steps when executing the program: Figure 5

[0157] Obtain a plurality of spatio-temporal subgraphs corresponding to a plurality of charging piles in a region, wherein one spatio-temporal subgraph is used to indicate at least two items of state data of one charging pile;

[0158] In a case where at least two spatio-temporal subgraphs meeting an expected condition are determined from the plurality of spatio-temporal subgraphs, combine the at least two spatio-temporal subgraphs to obtain a spatio-temporal subgraph cluster, wherein the expected condition is used to indicate that a similarity degree of the at least two spatio-temporal subgraphs is greater than a preset threshold;

[0159] Reconstruct each spatio-temporal subgraph in the spatio-temporal subgraph cluster, wherein the reconstructed spatio-temporal subgraph indicates more items of state data of the charging pile;

[0160] Use the reconstructed spatio-temporal subgraph cluster to predict the occupancy rate of the charging pile in the region.

[0161] The device herein can be a server, a PC, a PAD, a mobile phone and the like.

[0162] The present application also provides a computer program product, which is adapted to execute the program of the following method steps when executed on a data processing device:

[0163] Obtain a plurality of spatio-temporal subgraphs corresponding to a plurality of charging piles in a region, wherein one spatio-temporal subgraph is used to indicate at least two items of state data of one charging pile;

[0164] ​In a case where at least two spatio-temporal subgraphs meeting an expected condition are determined from the plurality of spatio-temporal subgraphs, the at least two spatio-temporal subgraphs are combined to obtain a spatio-temporal subgraph cluster, wherein the expected condition is used to indicate that a similarity degree of the at least two spatio-temporal subgraphs is greater than a preset threshold;

[0165] feature reconstruction is performed on each spatio-temporal subgraph in the spatio-temporal subgraph cluster, wherein the reconstructed spatio-temporal subgraph indicates more state data of the charging pile;

[0166] The reconstructed spatio-temporal subgraph cluster is used to perform occupancy prediction of the charging pile in the region.

[0167] Those skilled in the art will understand that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer usable program code.

[0168] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce an apparatus that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The functions specified in a flow or multiple flows and / or blocks

[0169] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The functions specified in a flow or multiple flows and / or blocks

[0170] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a process for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocksFigure 1 The steps of a method, process, or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module can reside in RAM, flash memory, ROM, or any other form of non- transitory computer-readable storage medium known in the art. Any of the noted components can be used in conjunction with the embodiments disclosed herein. It is understood that the systems and / or associated components can include appropriate software, firmware, and / or hardware to implement the embodiments disclosed herein.

[0171] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0172] The memory can include non-persistent memory and / or persistent memory, embodied as random access memory (RAM), read only memory (ROM), and / or flash memory, among others. The memory is an example of computer readable media.

[0173] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technologies, compact disc read only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassettes, magnetic storage devices or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media such as modulated data signals and carrier waves.

[0174] It is also important to note that the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0175] Those skilled in the art will appreciate that embodiments of the present application can be devised for a method, a system, or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer readable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code thereon.

[0176] The above merely provides an example of the present application, but is not intended to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the scope of claims of the present application.

Claims

1. A method for predicting the occupancy rate of charging piles, characterized in that, include: Obtain multiple spatiotemporal subgraphs corresponding to multiple charging piles within the area, wherein a spatiotemporal subgraph is used to indicate at least two state data of a charging pile; If at least two spatiotemporal subgraphs that meet the expected conditions are determined from the plurality of spatiotemporal subgraphs, the at least two spatiotemporal subgraphs are combined to obtain a spatiotemporal subgraph cluster, wherein the expected conditions are used to indicate that the similarity between the at least two spatiotemporal subgraphs is greater than a preset threshold. Feature reconstruction is performed on each spatiotemporal subgraph in the spatiotemporal subgraph cluster, wherein the reconstructed spatiotemporal subgraph indicates more status data of the charging pile; The occupancy rate of charging piles in the region is predicted using the reconstructed spatiotemporal subgraph cluster.

2. The method according to claim 1, characterized in that, Before combining the at least two spatiotemporal subgraphs to obtain a spatiotemporal subgraph cluster, the method further includes: The central spatiotemporal subgraph corresponding to the central charging pile is determined from the plurality of spatiotemporal subgraphs; Candidate spatiotemporal subgraphs with similarity greater than the preset threshold are determined from the plurality of spatiotemporal subgraphs; If the number of candidate spatiotemporal subgraphs is greater than or equal to a subgraph threshold, the central spatiotemporal subgraph and the subgraphs with the highest similarity among the candidate spatiotemporal subgraphs are identified as the spatiotemporal subgraph cluster.

3. The method according to claim 2, characterized in that, After determining the candidate spatiotemporal subgraphs with a similarity greater than the preset threshold from the plurality of spatiotemporal subgraphs, the method further includes: If the number of candidate spatiotemporal subgraphs is less than a subgraph threshold, the spatiotemporal subgraph with the highest similarity among the candidate spatiotemporal subgraphs is determined as a new central spatiotemporal subgraph, and a new candidate spatiotemporal subgraph with a similarity greater than the preset threshold is determined from the plurality of spatiotemporal subgraphs. At least one of the central spatiotemporal subgraphs, the candidate spatiotemporal subgraphs, and the new candidate spatiotemporal subgraphs is determined as the spatiotemporal subgraph cluster.

4. The method according to claim 1, characterized in that, The feature reconstruction of each spatiotemporal subgraph in the spatiotemporal subgraph cluster includes: According to a multi-channel sliding window with a preset step size, the original features of each spatiotemporal subgraph are recombined to obtain each spatiotemporal subgraph after feature recombination. The number of feature nodes corresponding to the spatiotemporal subgraph after feature recombination is N times the number of feature nodes of the spatiotemporal subgraph before feature recombination, where N is the number of channels of the multi-channel sliding window.

5. The method according to claim 1, characterized in that, The step of using the reconstructed spatiotemporal subgraph cluster to predict the occupancy rate of charging piles in the region includes: Based on each reconstructed spatiotemporal subgraph in the reconstructed spatiotemporal subgraph cluster, multiple prediction sub-models are trained for the multiple charging piles, wherein one prediction sub-model corresponds to one reconstructed spatiotemporal subgraph; The occupancy rate of the multiple charging piles in the area is predicted using the multiple prediction sub-models.

6. The method according to claim 5, characterized in that, The step of using the multiple prediction sub-models to predict the occupancy rate of the multiple charging piles in the area includes: In the process of using the multiple prediction sub-models to predict the occupancy rate of the target charging piles in the area, multiple prediction results obtained by the multiple prediction sub-models to predict the occupancy rate of the target charging piles are obtained, wherein the multiple charging piles include the target charging piles. The multiple prediction results are weighted and summed to obtain the occupancy rate prediction results of the multiple prediction sub-models for the target charging pile.

7. The method according to claim 5, characterized in that, The method further includes: According to a preset period, obtain a first number of reconstructed spatiotemporal subgraph clusters; Using the multiple prediction sub-models, based on each reconstructed spatiotemporal subgraph in the first number of reconstructed spatiotemporal subgraph clusters, the occupancy rate of the multiple charging piles in the region is predicted to obtain the occupancy rate prediction results of the multiple charging piles. The occupancy rate prediction result of a charging pile includes the occupancy rate prediction sub-results of a second number of cycles, where the second number is less than the first number.

8. A charging pile occupancy prediction device, characterized in that, include: The acquisition unit is used to acquire multiple spatiotemporal sub-graphs corresponding to multiple charging piles in the area, wherein a spatiotemporal sub-graph is used to indicate at least two state data of a charging pile; A combination unit is used to combine at least two spatiotemporal subgraphs to obtain a spatiotemporal subgraph cluster when at least two spatiotemporal subgraphs that meet the expected conditions are determined from multiple spatiotemporal subgraphs, wherein the expected conditions are used to indicate that the similarity between at least two spatiotemporal subgraphs is greater than a preset threshold. The reconstruction unit is used to reconstruct the features of each spatiotemporal subgraph in the spatiotemporal subgraph cluster. The reconstructed spatiotemporal subgraph indicates more status data of the charging pile. The prediction unit is used to predict the occupancy rate of charging piles in a region using the reconstructed spatiotemporal subgraph cluster.

9. A processor, characterized in that, The processor is used to run a program, wherein the program executes the method according to any one of claims 1 to 7 when it runs.

10. An electronic device, characterized in that, The method includes one or more processors and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of any one of claims 1 to 7.

Images