DE102022119607A1 - Computer-implemented method for predicting the availability of an infrastructure component and route planning based on the prediction - Google Patents

Abstract

In order to improve the use of the infrastructure or route planning, a prediction system (10) is proposed. The prediction system (10) is configured to predict an availability value (Yi) of infrastructure components within a road network represented by road network structure data, the system comprising: a) supplementary means (12) configured to provide historical availability data (Xi ) expand the infrastructure component and road network structure data with at least one attribute data matrix (α, β) to obtain attribute-enhanced availability data (Ei), the historical availability data (Xi) indicating an availability value of the infrastructure component at different points in time; b) a first machine learning model ( 14) including at least one graph convolutional network layer configured to determine spatial dependence output data indicative of the spatial dependence of the historical availability data (Xi) based on the attribute-enhanced availability data (Ei) and the road network structure data; c) a second Machine learning model (16) containing an informer layer trained to determine an availability score of the infrastructure component based on the spatial dependency output data, the informer layer containing at least one encoder (20) and at least one decoder (22). , wherein the encoder (20) determines a feature map (28) based on the spatial dependence output data, the feature map (28) being input to each decoder (22).

Description

TECHNICAL FIELD

The invention relates to a computer-implemented method for predicting the availability of an infrastructure component, such as. B. a charging station for electric vehicles.

BACKGROUND

Accurate traffic information prediction plays an important role in managing a smart city. Generally, traffic information includes connection speed, traffic flow, vehicle density, travel time, facility usage status, etc. With the rapid development of electric vehicle EV, the proportion of electric vehicles is increasing. However, the limited range and the limited number of charging stations as well as the significantly longer charging times compared to the short refueling time of petrol cars lead to range anxiety among drivers of electric vehicles. As one of the most important infrastructure for electric vehicles, charging stations have recently attracted more attention. Some studies show that the charging behavior of electric vehicles has a certain regularity. Therefore, an accurate EV charging station usage prediction system can effectively reduce range anxiety and improve road traffic efficiency.

Another advantage of the large number of intelligent sensors is real-time monitoring at the station level. Most canonical methods for predicting facility usage status rely on historical traffic characteristics to make predictions. However, predicting the availability of EV charging stations is generally much more complex than other time series predictions because future availability not only depends on past availability values, but is also associated with topological relationships or extensive external influences. For example, using a charging station on campus or in a central business district can be heavily influenced by commute time. In contrast to the availability in a residential area, a significant increase in availability can be observed during off-duty times, even if the street structure may be similar. Another example is that bad weather, such as heavy rain, can limit commutes and delay commute times, which in turn impacts charging station usage.

In the known methods, taking into account the randomness caused by these external factors is a major challenge. With the progressive development of deep learning technologies, several prediction methods have been proposed to solve this problem, e.g. B. the ARIMA method (Auto regressive and Integrated Moving Average), the CNN method (Convolutional Neural Network), the LSTM method (Long Short-Term Memory), the GCN method (Graph Convolution Network) and transformer-based methods. Each algorithm has its own strengths and limitations. Most of the well-known models do not take external factors into account and, accordingly, have a weaker perception of these external factors.

Additionally, reference is made to the following documents:

• [1] R. Luo and R. Su, "Traffic signal transition time prediction based on aerial captures during peak hours," in 2020 16th International Conference on Control, Automation, Robotics and Vision (ICARCV). IEEE, 2020, pp. 104-110 .
• [2] S. Storandt and S. Funke, “Cruising with a battery-powered vehicle and not getting stranded,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 26, no. 1 , 2012.
• [3] K. Qian, C. Zhou, M. Allan, and Y. Yuan, "Load model for prediction of electric vehicle charging demand," in 2010 International Conference on Power System Technology. IEEE, 2010, pp. 1-6 .
• [4] Y. Xiong, J. Gan, B. An, C. Miao, and AL Bazzan, "Optimal electric vehicle fast charging station placement based on game theoretical framework," IEEE Transactions on Intelligent Transportation Systems, vol. 19, no. 8, pp. 2493-2504 , 2017.
• [5] G. Alface, JC Ferreira, and R. Pereira, “Electric vehicle charging process and parking guidance app,” Energies, vol. 12, no. 11, p. 2123 , 2019.
• [6] J. Zhu, Q. Wang, C. Tao, H. Deng, L. Zhao, and H. Li, "Ast-gcn: Attribute augmented spatiotemporal graph convolutional network for traffic forecasting," IEEE Access, vol. 9, pp. 35 973-35 983 , 2021.
• [7] C.-W. Huang, C.-T. Chiang, and Q. Li, “A study of deep learning networks on mobile traffic forecasting,” in 2017 IEEE 28th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC). IEEE, 2017, pp. 1-6 .
• [8th] S. Siami-Namini, N. Tavakoli, and AS Namin, “A comparison of arima and Istm in forecasting time series,” in 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA). IEEE, 2018, pp. 1394-1401 .
• [9] W. Jiang and J. Luo, “Graph neural network for traffic forecasting: A survey,” arXiv preprint arXiv:2101.11174, 2021.
• [10] J. Yang, H. Chen, Y. Xu, Z. Shi, R. Luo, L. Xie, and R. Su, "Domain adaptation for degraded remote scene classification," in 2020 16th International Conference on Control, Automation, Robotics and Vision (ICARCV). IEEE, 2020, pp. 111-117 .
• [11] H. Zhao, H. Yang, Y. Wang, D. Wang, and R. Su, "Attention based graph bi-Istm networks for traffic forecasting," in 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2020, pp. 1-6 .
• [12] X. Luo, D. Li, and S. Zhang, “Traffic flow prediction during the holidays based on dft and svr,” Journal of Sensors, vol. 2019, 2019.
• [13] H. Zhao, H. Yang, Y. Wang, R. Su, and D. Wang, "Domain-adversarial based temporal graph convolutional network for traffic flow prediction problem," in 2021 IEEE International Intelligent Transportation Systems Conference (ITSC). IEEE, 2021, pp. 1365-1370 .
• [14] H. Yang, Y. Wang, H. Zhao, J. Zhu, and D. Wang, "Real-time traffic incident detection using an autoencoder model," in 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2020, pp. 1-6 .
• [15] D. Zhang and M.R. Kabuka, "Combining weather condition data to predict traffic flow: A gru-based deep learning approach," IET Intelligent Transport Systems, vol. 12, no. 7, pp. 578-585 , 2018.
• [16] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, AN Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” in Advances in neural information processing systems, 2017, pp. 5998-6008 .
• [17] R. Luo, Y. Zhang, Y. Zhou, H. Chen, L. Yang, J. Yang, and R. Su, “Deep learning approach for long-term prediction of electric vehicle (ev) charging station availability,” in 2021 IEEE International Intelligent Transportation Systems Conference (ITSC). IEEE, 2021, pp. 3334-3339 .
• [18] B. Liao, J. Zhang, C. Wu, D. Mcllwraith, T. Chen, S. Yang, Y. Guo, and F. Wu, "Deep sequence learning with auxiliary information for traffic prediction," in Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2018, pp. 537-546 .
• [19] L. Zhao, Y. Song, C. Zhang, Y. Liu, P. Wang, T. Lin, M. Deng, and H. Li, "T-gcn: A temporal graph convolutional network for traffic prediction," IEEE Transactions on Intelligent Transportation Systems, vol. 21, no. 9, pp. 3848-3858 , 2019.
• [20] T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” arXiv preprint arXiv:1609.02907, 2016.
• [21] Y.-H. H. Tsai, S. Bai, M. Yamada, L.-P. Morency, and R. Salakhutdinov, “Transformer dissection: A unified understanding of transformer's attention via the lens of kernel,” arXiv preprint arXiv:1908.11775, 2019.
• [22] H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, and W. Zhang, “Informer: Beyond efficient transformer for long sequence time-series forecasting,” in Proceedings of AAAI, 2021.

BRIEF SUMMARY OF THE INVENTION

It is the object of the invention to improve the use of the infrastructure by vehicles. This task is solved by the subject matter of the independent claims. Preferred embodiments are the subject of the dependent claims.

1. a) Ergänzen von historischen Verfügbarkeitsdaten der Infrastrukturkomponente und von Straßennetzstrukturdaten mit mindestens einer Attributdatenmatrix, um attribut-erweiterte Verfügbarkeitsdaten zu erhalten, wobei die historischen Verfügbarkeitsdaten einen Verfügbarkeitswert der Infrastrukturkomponente zu verschiedenen Zeitpunkten anzeigen;
2. b) Einspeisen der attribut-ergänzten Verfügbarkeitsdaten und der Straßennetzstrukturdaten in ein erstes maschinelles Lernmodell, das mindestens eine Graphfaltungsnetzwerkschicht enthält, die zum Ermitteln von Raumabhängigkeitsausgabedaten konfiguriert ist, die die räumliche Abhängigkeit der historischen Verfügbarkeitsdaten angeben;
3. c) Einspeisen der Raumabhängigkeitsausgabedaten in ein zweites maschinelles Lernmodell, das eine Informer-Schicht enthält, die zum Ermitteln einer Verfügbarkeitsbewertung der Infrastrukturkomponente trainiert ist, wobei die Informer-Schicht mindestens einen Encoder und mindestens einen Decoder enthält, wobei der Encoder eine Featurekarte basierend auf den Raumabhängigkeitsausgabedaten bestimmt, wobei die Featurekarte in jeden Decoder eingespeist wird.

The invention provides a computer-implemented method for predicting an availability rating of infrastructure components within a road network represented by road network structure data, the method comprising:

1. a) supplementing historical availability data of the infrastructure component and road network structure data with at least one attribute data matrix to obtain attribute-enhanced availability data, wherein the historical availability data indicates an availability value of the infrastructure component at different points in time;
2. b) feeding the attribute-enhanced availability data and the road network structure data into a first machine learning model that includes at least one graph convolutional network layer configured to determine spatial dependence output data indicative of the spatial dependence of the historical availability data;
3. c) feeding the spatial dependency output data into a second machine learning model that includes an informer layer trained to determine an availability score of the infrastructure component, wherein the Informer layer includes at least one encoder and at least one decoder, the encoder determining a feature map based on the spatial dependence output data, the feature map being fed to each decoder.

Preferably, in step a), a first attribute data matrix represents static external factors that are associated with each infrastructure component, the static external factors being determined by time-independent properties of the environment of the respective infrastructure component.

Preferably, in step a), a second attribute data matrix represents dynamic external factors associated with each infrastructure component and/or the road network, wherein the dynamic external factors are determined by time-dependent events that occur in the environment in which the respective infrastructure component is arranged.

Preferably, a time window of a certain length is selected, and only events within this time window are added to the historical availability data.

Preferably, the informer layer in step c) contains a first encoder and a last encoder, the output of step b) being fed into the first encoder and the output of the first encoder being fed into a further encoder, the feature map from the last encoder is output.

Preferably, the informer layer in step c) contains a first decoder and a last decoder, the output of the last decoder being fed into a fully connected output layer.

Preferably, in step c), the spatial dependence output data is sequentially partially masked, and the partially sequentially masked data is fed into the decoder or the first decoder.

Preferably, in step c), each encoder contains a first self-attention pyramid and a second self-attention pyramid, the second self-attention pyramid having fewer inputs than the first self-attention pyramid and each self-attention pyramid creates a semi-feature map that is chained together to form the initial feature map.

Preferably, the second self-attention pyramid has half of the inputs of the first self-attention pyramid.

1. a) Ermitteln einer anfänglichen Route von einer Startposition zu einem Ziel;
2. b) Ermitteln einer Streckenlänge der anfänglichen Route und Vergleichen der Routenlänge mit einer verbleibenden Entfernung, die das Fahrzeug aufgrund des Kraftstoffverbrauchs noch zurücklegen kann;
3. c) wenn festgestellt wird, dass die Routenlänge länger ist als die verbleibende Entfernung, Ausführen eines zuvor beschriebenen Verfahrens, um mindestens eine Verfügbarkeitsbewertung zu erhalten, Ermitteln eines Zwischenstopps auf der Grundlage der Verfügbarkeitsbewertung und Ändern der ursprünglichen Route, um den Zwischenstopp einzuschließen und eine endgültige Route zu erhalten;
4. d) Erzeugung eines Steuersignals, das das Fahrzeug veranlasst, einen Fahrer über die endgültige Route zu informieren, oder das das Fahrzeug veranlasst, der endgültigen Route zu folgen.

The invention provides a computer-aided method for planning a route of a vehicle, the method comprising:

1. a) determining an initial route from a starting position to a destination;
2. b) determining a route length of the initial route and comparing the route length with a remaining distance that the vehicle can still travel based on fuel consumption;
3. c) if it is determined that the route length is longer than the remaining distance, performing a previously described procedure to obtain at least one availability score, determining a stopover based on the availability score and changing the original route to include the stopover and a final one route to get;
4. d) generating a control signal that causes the vehicle to inform a driver of the final route or that causes the vehicle to follow the final route.

The invention includes a computer program that contains instructions that, when executed by a computer, cause the computer to perform one, some, or all of the steps of a preceding method.

The invention includes a computer-readable storage medium or a data carrier signal that contains the computer program.

1. a) Ergänzungsmittel I, die zum Ergänzen von historischen Verfügbarkeitsdaten der Infrastrukturkomponente und von Straßennetzstrukturdaten mit mindestens einer Attributdatenmatrix konfiguriert sind, um attribut-ergänzte Verfügbarkeitsdaten zu erhalten, wobei die historischen Verfügbarkeitsdaten einen Verfügbarkeitswert der Infrastrukturkomponente zu verschiedenen Zeitpunkten angeben;
2. b) ein erstes maschinelles Lernmodell, das mindestens eine Graphfaltungsnetzwerkschicht enthält, die zum Ermitteln von Raumunabhängigkeitsausgangsdaten konfiguriert ist, die die räumliche Abhängigkeit der historischen Verfügbarkeitsdaten auf der Grundlage der attribut-erweiterte Verfügbarkeitsdaten und der Straßennetzstrukturdaten angeben;
3. c) ein zweites maschinelles Lernmodell, das eine Informer-Schicht enthält, die zum Ermitteln der Verfügbarkeitsbewertung der Infrakstrukturkomponente auf der Grundlage der Raumunabhängigkeitsausgangsdaten trainiert ist, wobei die Informer-Schicht mindestens einen Encoder und mindestens einen Decoder enthält, wobei der Encoder eine Featurekarte auf der Grundlage der Raumunabhängigkeitsausgangsdaten bestimmt, wobei die Featurekarte in jeden Decoder eingespeist wird. Die Erfindung stellt eine Fahrzeugroutenplanungsvorrichtung für ein Fahrzeug bereit, die ein Vorhersagesystem umfasst.

The invention provides a prediction system configured for predicting an availability value of infrastructure components within a road network represented by road network structure data, the system comprising:

1. a) supplementary means I, which are configured to supplement historical availability data of the infrastructure component and road network structure data with at least one attribute data matrix in order to obtain attribute-supplemented availability data, the historical availability data indicating an availability value of the infrastructure component at different points in time;
2. b) a first machine learning model that includes at least one graph convolutional network layer configured to determine spatial independence output data that shows the spatial dependence of the historical availability data based on the attribute provide extended availability data and road network structure data;
3. c) a second machine learning model that includes an informer layer trained to determine the availability score of the infrastructure component based on the spatial independence output data, the informer layer including at least one encoder and at least one decoder, the encoder a feature map on the Based on the spatial independence output data, the feature map is fed into each decoder. The invention provides a vehicle route planning device for a vehicle that includes a prediction system.

One idea is to build a neural network capable of extracting both spatial-temporal information and external influences to predict the charging station's usage status, i.e. availability. An Attribute-Augmented Spatial-Temporal Graph Informer Network (AST-GIN) is used as a machine learning model to determine the availability value. The AST-GIN includes an Attribute Augmentation Unit (A2U), a Graph Convolutional Network (GCN), and an Informer Network. The disclosed AST-GIN model, unlike known models, can well consider the influence of dynamic and static external factors on the EV charging station. As described later, the AST-GIN is compared to known baseline models by testing it on a real data set collected in Dundee.

There are different approaches to traffic prediction and the methods can generally be divided into two types: canonical models and deep learning-based models. Canonical prediction models are usually based on mathematical models and treat traffic behavior as a conditional process. There are many well-known models, such as the Historical Average (HA) model, the K-nearest neighbor model, the ARIMA model, and the Support Vector Regression (SVR) model.

Most of these models take into account the trend of the data and assume that time series data is stable, which makes it difficult for these models to respond to a rapid change in inputs. Recently, deep learning-based prediction methods have been widely used for time series prediction. These models are typically capable of extracting nonlinear relationships from the input sequence and include a recurrent neural network (RNN), a stacked autoencoding neural network (SAE), a gated recurrent unit (GRU ), an LSTM model (Long Short-Term Memory), a Transformer and variants thereof. The models are more efficient at extracting the temporal information than canonical prediction models.

As already described, external factors influence the future availability of charging stations for EVs. Existing procedures can be further improved by taking external information flows into account. This is done with the AST-GIN for charging station availability prediction, which integrates both spatiotemporal and external factors as inputs to improve the model's perceptive ability during prediction.

The aim of the invention is to predict the future availability of an EV charging station based on historical conditions and associated information. Based on the introduced a priori knowledge, the demand for EVs has a strong periodicity and external factors have a high correlation with the usage of EVs.

The chance of finding an availability function that estimates an availability value for a charging station is very promising. As an example, historical aggregated charging station availability data and two external factors, weather conditions and Point Of Interest (POI) information, are used to illustrate the idea.

The road network connecting the different charging stations is treated as a weighted undirected graph G = {V, E}. The charging stations are represented by the nodes V within the graph. E is the edge set of the graph representing the connectivity between the stations. The corresponding adjacency matrix A can be constructed based on node and edge information. Using a road map based on the latitude and longitude of the charging stations, the distance between EV stations can be estimated. The elements of the adjacency matrix are preferably calculated using a Gaussian kernel weighting function that has a predetermined limit. In other words: Only if the distance between two stations is smaller than a specified limit distance is the matrix element greater than 0.

Another element is a traffic feature matrix X _i , which contains high-dimensional information about the availability of charging stations. In other words: the traffic feature matrix contains the overall historical availability rating of each charging station being modeled.

Thus, at time i, the known historical states of the L time steps [X _i - _L , X _i-L+1 , ...,X _i ] are used as partial inputs to determine the states of the next M steps [Y _i+1 , ...,Y _i+M ] to predict.

Furthermore, the influence of external factors is considered as the associated attribute matrix F. These factors form an attribute matrix [F ₁ , F ₂ , ..., F _I ], where I is the dimension of the attribute information. At time i, the set of the jth associated information is F _j = [j _i+L ,j _i-L+1 , ..., j _i ].

The invention solves the technical problem of charging station availability prediction for EVs by considering external factors and refining the relationship function f based on the historical usage data X, the attribute matrix F and the road graph structure G to obtain the future usage values Y.

The AST-GIN model contains an Attribute Augmentation Unit (A2Unit) that can integrate the external information, a GCN layer and an Informer layer.

The historical time series data and the external data are fed into the A2Unit for attribute supplementation. Then the processed information is fed into the GCN layer to extract spatial information. Finally, the Informer layer adopts the outputs of the GCN layer to extract the temporal dependencies.

In order to comprehensively take the influence of external factors into account, dynamic and static attributes are selected for the target region. The historical availability matrix X of the EV stations, the road structure G and two types of attribute matrices are integrated into the A2Unit for expansion. A static attribute matrix α contains p categories of attributes that are time invariant. This may refer to the static environment of the charging station, e.g. B. whether there is a supermarket or other offers that you can use while the car is charging.

Similarly, β represents w different dynamic attributes that change over time. This includes dynamic environmental phenomena such as the weather, the traffic situation, etc. In order to aggregate the cumulative influence of the dynamic attributes, a historical window of length L is chosen. So the final augmented matrix processed by A2Unit at time i is E _i = [Xi, α, β , β _i-Li-L+1 , ... β _i ], and the same processing procedure is used for each timestamp applied within the traffic feature matrix X. In other words, for each time point i, one obtains the augmented matrix E _i by appending additional columns to the matrix X _i , which contain the static attributes α and a historical time window of the dynamic attributes β.

The GCN layer is used to extract the spatial dependence of the input data. The convolution is given by H _I+1 = σ(D ^-1/2 ÃD ^-1/2 H _I W _I ), where σ is the activation function, D is the diagonal degree matrix, Ä is the adjacency matrix with self-loops, W _{I is} the weight matrix of the Convolutional layer I, H _I+1 the output of the convolution layer, H _I the input, which is the extended matrix E for H ₀ .

The Informer layer is used to obtain global temporal dependence in prediction. The informer layer uses an encoder-decoder architecture. The model improves the efficiency of query attention by measuring sparsity and reducing the dimension of the query value in the Multi-head ProbSparse Self-attention block.

BRIEF DESCRIPTION OF THE DRAWINGS

• 1 zeigt eine Ausführungsform eines AST-GIN-Modells;
• 2 zeigt eine Ausführungsform einer Informer-Schicht;
• 3 zeigt eine Tabelle mit den Versuchsergebnissen; und
• 4 zeigt ein Diagramm zur Veranschaulichung der Genauigkeit der verglichenen Modelle.

Embodiments of the invention are described in more detail with reference to the accompanying schematic drawings.

• 1 shows an embodiment of an AST-GIN model;
• 2 shows an embodiment of an informer layer;
• 3 shows a table with the experimental results; and
• 4 shows a diagram to illustrate the accuracy of the compared models.

DETAILED DESCRIPTION OF THE EMBODIMENTS

In 1 a prediction system 10 is partially shown. The prediction system 10 is configured to provide an availability value of an infrastructure component, such as. B. an electric charging station for electric vehicles. Typically, several charging stations are distributed throughout an area and connected by a road network. It should be noted that the invention is described based on electrical charging infrastructure, but the ideas described herein can be readily applied to other infrastructure components.

Each charging station is located in an environment that may have static and/or dynamic characteristics. Static attributes are time-independent, e.g. B. surrounding shops or offers, while dynamic attributes, e.g. B. the weather changes over time.

The prediction system 10 is fed with historical availability data X _i for each charging station. The historical availability data can provide information about how many individual chargers are available at the charging station.

The prediction system 10 is fed with a static attribute data matrix α. The static attribute data matrix α is an indicator of the static environment around the electric charger.

The prediction system 10 is fed with a dynamic attribute data matrix β containing entries for each time i in the past for a selectable time window L. For example, the dynamic attribute matrix β contains weather data for each charging station for the last L hours.

The prediction system 10 includes a supplement 12. The supplement 12 receives the historical availability data X _i and the attribute data matrices α, β. The supplementary means 12 concatenates this input data into attribute-supplemented availability data E _i .

The prediction system 10 includes a first machine learning model 14 with at least one graph convolutional network (GCN) layer. The road network connecting the charging stations is represented as an undirected graph G. Each charging station is represented by a node V and the road network is represented by the edges E. A Gaussian kernel function with a specified cut-off value is used for the adjacency matrix A. The GCN extracts the spatial dependence of the input data and feeds the result into a second machine learning model 16.

As in 2 shown, the prediction system 10 includes the second machine learning model 16, which contains at least one informer layer 18. The informer layer 18 has at least one encoder 20 and at least one decoder 22.

The encoder 20 includes a first self-attention pyramid 24 and a second self-attention pyramid 26, which has half of the inputs of the first self-attention pyramid 24. The two self-attention pyramids 24 and 26 are configured as a ProbSparse self-attention pyramid. The outputs of the self-attention pyramid 24 and 26 are combined into a linked feature map 28.

The concatenated feature map 28 is fed to the decoder 22 in a multi-head attention block 30. The multi-head attention block 30 receives the output of a masked self-attention pyramid 32 which is fed with masked availability data 34. The decoder 22 has an output layer that normalizes the decoder output to output an availability value for each charging station.

The final output of the forecasting system 10 is an expected time series of availability values expected for each point i within a predetermined future time frame, e.g. B. from 30 minutes to 120 minutes in 30-minute increments, contains a prediction of the availability values Y _i of each charging station in the road network.

This output can be fed into a route planning system that is known and capable of planning a route based on the remaining distance of a battery charge. In addition, the route is also determined based on the availability rating Y _i of the charging stations near the route.

The route planning system can determine a first route based on user input and/or position data. The load required to travel along the route is then determined. If the remaining energy for the remaining route in the vehicle's energy storage is insufficient to reach the destination, the route is changed to include at least one charging station based on the predicted availability value Y _i at the time of arrival at the respective charging station . In this way, the electrical infrastructure and available resources can be used more efficiently. In addition, drivers' range anxiety can be reduced.

To evaluate the performance of the AST-GIN model, the necessary experiments were conducted on the EV charging station availability dataset. 5 efficient time series forecasting base models were selected for comparison. During the experiment, the performance of the AST-GIN model is evaluated with only static external factors, only with dynamic external factors, and with both static and dynamic factors separately.

Dundee EV charging dataset: This dataset (publicly available on the website) is a record of EV charging behavior in Dundee, Scotland. There are 57 charging points in Dundee which can be divided into 3 types including Slowcharger, Fastcharger and Rapidcharger. A total of 3 valid data sets are available, which can be found in 3 different time periods: 01/09/17 to 01/12/17, 02/12/17 to 02/03/18 and 05/03/18 to 05/06/18. The geographical location of all charging stations is also indicated.

The data set used for the experiment was recorded between 05/03/18 and 05/06/18. A total of 16773 charging sessions were recorded, and each of the charging session records includes the charging station ID, charging port ID, charging start and end time, total electric power consumed, geographical location and charging station type. There are 40 slow chargers with 5894 recorded charging processes, 8 fast chargers with 1416 recorded charging processes and 9 rapid chargers with 9463 recorded charging processes.

Additionally, the Dundee weather dataset available on the website is a weather record for the city of Dundee. The weather is recorded hourly, and each record contains the general description of the weather, temperature, wind, humidity and barometer.

wobei Σ t_p die gesamte Ladedauer innerhalb einer bestimmten Sitzung und M_connector die Anzahl der Ladestecker an dieser Station ist.The availability of a specific charging station in the p-th session is calculated every 30 minutes and can be described as follows ${\alpha }_p=1-\frac{\sum t_p}{30\cdot M_{\text{connector}}},$

where Σ t _p is the total charging time within a specific session and M _connector is the number of charging connectors at this station.

In the experiment, the charging station environment is divided into 8 categories, including transportation services, catering services, shopping services, educational services, accommodation services, medical services, residential services and others. The category of environment with the largest share is called the POI value of a charging station based on its geographical location.

The weather in Dundee is classified into 5 types: sunny, cloudy, foggy, slightly rainy and heavily rainy with different labels from 1 to 5. Since the time interval of the weather data from the source dataset is one hour, the weather in the recorded period is considered to be the same , d. i.e., if the weather at 5:50 p.m. is recorded as sunny, the weather at 6:20 p.m. would be recorded as sunny.

A 3-layer GCN structure is used in the experiment. For each informer block, the number of encoder layers is 2 and the number of decoder layers is 3. During training, 50% of the data is randomly divided for training, 33% of the data is selected for evaluation, and the rest 17% of the data is used for the test. The network is optimized using the Adam optimizer. The learning rate starts at e ^-4 and decreases by a factor of 10 every 2 epochs. The total number of epochs is 50 with an early stopping provision. The batch size is chosen to be 32 and the loss function is the mean squared error. The entire network is trained on a GPU RTX3060.

Common metrics for evaluating the predictive performance of the models were selected for the experiment, namely Root Mean Squared Error (RMSE), Coefficient of Determination (R ² ), Explained Variance Score (var), Mean Absolute Error (MAE) and Accuracy .

Five prior art basic models were used to compare performance with the inventive AST-GIN model. The basic models include GRU, LSTM, Transformer, Informer and Spatial Temporal Transformer Network (STTN). Based on the 30 minute time interval of EV charging availability data, the selected models are deployed to predict availability in the next 30, 60, 90 and 120 minute horizons. The numerical results are in 3 shown.

When predicting the availability of charging stations in the short term, e.g. B. with a time horizon of 30 minutes, the AST-GIN model effectively captures the temporal dependence of the data and outperforms the other basic models with higher accuracy. For the 90-minute forecast, the AST-GIN model with the dynamic external factors achieves an accuracy of 0.8388, while the best model, STTN, achieves an accuracy of 0.7520 among the base models. AST-GIN outperforms STTN on the 90 minute horizon with 11.54% higher accuracy. For long-term forecasting, e.g. 120 minutes, AST-GIN still has the best performance with the highest accuracy compared to the other base models. The accuracy of AST-GIN, 0.7507, is 8.97% higher than that of the best model STTN among the baseline models. The consolidated result is in 4 shown.

Among the three types of external factors used, which include only static factors, only dynamic factors and the combination of both factors, the use of the combination of external factors generally leads to better performance.

In the present application, the deep learning model AST-GIN is used to predict the availability of EV charging stations, taking into account the influence of external factors described and verified. The model includes an A2Unit layer, at least one GCN layer and at least one Informer layer to complement time-serial traffic features and extract spatio-temporal dependencies of an EV charging station usage state. AST-GIN and the baselines are tested using data from the city of Dundee. The experiments show that AST-GIN has better prediction capabilities across different time horizons and metrics. In summary, AST-GIN is able to effectively consider the influence of comprehensive external attributes and predict the usage state of EV charging stations. AST-GIN can be applied to EV charging system applications to accurately predict charging demand and supply and suggest better charging behavior to drivers, or to provide key information for a planning system or route planning system.

REFERENCE MARKS

10

Prediction system

12

Supplements

14

first machine learning model

16

second machine learning model

18

Informer layer

20

Encoder

22

decoder

24

first self-attention pyramid

26

second self-attention pyramid

28

concatenated feature map

30

Multi-head attention block

32

masked self-attention pyramid

34

masked availability data

A

Adjacency matrix

E

edge

egg

attribute-enhanced availability data

G

undirected graph

v

node

Xi

historical availability data

α

static attribute data matrix

β

dynamic attribute data matrix

QUOTES INCLUDED IN THE DESCRIPTION

This list of documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Non-patent literature cited

• R. Luo and R. Su, "Traffic signal transition time prediction based on aerial captures during peak hours," in 2020 16th International Conference on Control, Automation, Robotics and Vision (ICARCV). IEEE, 2020, pp. 104-110 [0004]
• S. Storandt and S. Funke, “Cruising with a battery-powered vehicle and not getting stranded,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 26, no. 1 [0004]
• K. Qian, C. Zhou, M. Allan, and Y. Yuan, "Load model for prediction of electric vehicle charging demand," in 2010 International Conference on Power System Technology. IEEE, 2010, pp. 1-6 [0004]
• Y. Xiong, J. Gan, B. An, C. Miao, and A. L. Bazzan, "Optimal electric vehicle fast charging station placement based on game theoretical framework," IEEE Transactions on Intelligent Transportation Systems, vol. 19, no. 8, pp. 2493-2504 [0004]
• G. Alface, J. C. Ferreira, and R. Pereira, “Electric vehicle charging process and parking guidance app,” Energies, vol. 12, no. 11, p. 2123 [0004]
• J. Zhu, Q. Wang, C. Tao, H. Deng, L. Zhao, and H. Li, "Ast-gcn: Attribute augmented spatiotemporal graph convolutional network for traffic forecasting," IEEE Access, vol. 9, pp. 35 973-35 983 [0004]
• C.-W. Huang, C.-T. Chiang, and Q. Li, “A study of deep learning networks on mobile traffic forecasting,” in 2017 IEEE 28th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC). IEEE, 2017, pp. 1-6 [0004]
• S. Siami-Namini, N. Tavakoli, and A. S. Namin, “A comparison of arima and Istm in forecasting time series,” in 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA). IEEE, 2018, pp. 1394-1401 [0004]
• J. Yang, H. Chen, Y. Xu, Z. Shi, R. Luo, L. Xie, and R. Su, "Domain adaptation for degraded remote scene classification," in 2020 16th International Conference on Control, Automation, Robotics and Vision (ICARCV). IEEE, 2020, pp. 111-117 [0004]
• H. Zhao, H. Yang, Y. Wang, D. Wang, and R. Su, "Attention based graph bi-Istm networks for traffic forecasting," in 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2020, pp. 1-6 [0004]
• H. Zhao, H. Yang, Y. Wang, R. Su, and D. Wang, "Domain-adversarial based temporal graph convolutional network for traffic flow prediction problem," in 2021 IEEE International Intelligent Transportation Systems Conference (ITSC). IEEE, 2021, pp. 1365-1370 [0004]
• H. Yang, Y. Wang, H. Zhao, J. Zhu, and D. Wang, "Real-time traffic incident detection using an autoencoder model," in 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2020, pp. 1-6 [0004]
• D. Zhang and M. R. Kabuka, "Combining weather condition data to predict traffic flow: A gru-based deep learning approach," IET Intelligent Transport Systems, vol. 12, no. 7, pp. 578-585 [0004]
• A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” in Advances in neural information processing systems, 2017, pp. 5998-6008 [0004]
• R. Luo, Y. Zhang, Y. Zhou, H. Chen, L. Yang, J. Yang, and R. Su, “Deep learning approach for long-term prediction of electric vehicle (ev) charging station availability,” in 2021 IEEE International Intelligent Transportation Systems Conference (ITSC). IEEE, 2021, pp. 3334-3339 [0004]
• B. Liao, J. Zhang, C. Wu, D. Mcllwraith, T. Chen, S. Yang, Y. Guo, and F. Wu, "Deep sequence learning with auxiliary information for traffic prediction," in Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2018, pp. 537-546 [0004]
• L. Zhao, Y. Song, C. Zhang, Y. Liu, P. Wang, T. Lin, M. Deng, and H. Li, "T-gcn: A temporal graph convolutional network for traffic prediction," IEEE Transactions on Intelligent Transportation Systems, vol. 21, no. 9, pp. 3848-3858 [0004]

Claims (14)

Computer-implemented method for predicting an availability rating of infrastructure components within a road network represented by road network structure data, the method comprising: a) supplementing historical availability data (X _i ) of the infrastructure component and road network structure data with at least one attribute data matrix to produce attribute-supplemented availability data (E _i ), where the historical availability data (X _i ) indicates an availability value of the infrastructure component at different points in time; b) feeding the attribute-enhanced availability data (Ei) and the road network structure data into a first machine learning model (14) that contains at least one graph convolution network layer configured to determine spatial dependence output data that indicates the spatial dependence of the historical availability data (X _i ); c) feeding the spatial dependency output data into a second machine learning model (16), which contains an informer layer (18) which is trained to determine an availability assessment of the infrastructure component, the informer layer (18) having at least one encoder (20) and at least a decoder (22), the encoder (20) determining a feature map based on the spatial dependence output data, the feature map being fed to each decoder (22). Procedure according to Claim 1 , characterized in that in step a) a first attribute data matrix represents static external factors that are assigned to each infrastructure component, the static external factors being determined by time-independent properties of the environment of the respective infrastructure component. Method according to one of the preceding claims, characterized in that in step a) a second attribute data matrix represents dynamic external factors associated with each infrastructure component and/or the road network, the dynamic external factors being determined by time-dependent events occurring in the environment , in which the respective infrastructure component is arranged. Procedure according to Claim 3 , characterized in that a time window of a predetermined length is selected and only events within this time window are added to the historical availability data (X _i ). Method according to one of the preceding claims, characterized in that in step c) the informer layer (18) comprises a first encoder (20) and a last encoder (20), the output of step b) being in the first encoder (20 ) is fed and the output of the first encoder (20) is fed into a further encoder (20), the feature map being output by the last encoder (20). Method according to one of the preceding claims, characterized in that in step c) the informer layer (18) contains a first decoder (22) and a last decoder (22), the output of the last decoder (22) being converted into a fully connected Output layer is fed. Method according to one of the preceding claims, characterized in that in step c) the spatial dependence output data are successively partially masked and the successively partially masked data are fed into the decoder (22) or the first decoder (22). Method according to one of the preceding claims, characterized in that in step c) each encoder contains a first self-attention pyramid (24) and a second self-attention pyramid (26), the second self-attention pyramid (26 ) has fewer inputs than the first self-attention pyramid (24) and each self-attention pyramid (26) creates a semi-feature map that is linked together to form the output feature map. Procedure according to Claim 8 , characterized in that the second self-attention pyramid (26) has half of the inputs of the first self-attention pyramid (24). A computer-implemented method for planning a route of a vehicle, the method comprising: a) determining an initial route from a starting position to a destination; b) determining a route length of the initial route and comparing the route length with a remaining distance that the vehicle can travel based on fuel consumption; c) if it is determined that the route length is longer than the remaining distance, performing a method according to any one of the preceding claims to obtain at least one availability score, determining a stopover based on the availability score and modifying the original route to include the stopover to get an end to get valid route; d) generating a control signal that causes the vehicle to inform a driver of the final route or that causes the vehicle to follow the final route. Computer program containing instructions which, when executed by a computer, cause the computer to perform one, some or all of the steps of a method according to any one of the preceding claims. A computer-readable storage medium or a data carrier signal that the computer program recreates Claim 11 contains. Prediction system (10) configured to predict an availability rating of infrastructure components within a road network represented by road network structure data, the system comprising: a) supplementation means (12) suitable for supplementing historical availability data (X _i ) of the infrastructure component and of road network structure data are configured with at least one attribute data matrix in order to obtain attribute-supplemented availability data (E _i ), the historical availability data (X _i ) indicating an availability value of the infrastructure component at different points in time; b) a first machine learning model (14) containing at least one graph convolutional network layer configured to determine spatial dependence output data that determines the spatial dependence of the historical availability data (X _i ) based on the attribute-enhanced availability data (E _i ) and the road network structure data indicate; c) a second machine learning model (16) which contains an informer layer (18) which is trained to determine an availability assessment of the infrastructure component based on the spatial dependence output data, the informer layer (18) having at least one encoder (20) and contains at least one decoder (22), the encoder (20) determining a feature map based on the spatial dependence output data, the feature map being fed to each decoder (22). Vehicle route planning device for a vehicle, the device comprising a prediction system (10). Claim 13 includes.

Images