EP3816960B1 - Method and apparatus for outputting alert messages - Google Patents

Description

The present invention relates to a method for forecasting available parking spaces, in particular the forecast of non-commercial vehicle parking spaces, or "on-street" parking spaces, in urban traffic areas based on mobile phone data and other data sets. In addition, the invention also relates to an arrangement for predicting available parking spaces, which is designed to carry out the method.

Due to increasing restrictions and management, as well as increased traffic volume, non-commercial car parking spaces, the so-called "on-street" parking spaces, have become a scarce resource in inner cities. Traffic searching for parking pollutes the environment and affects the quality of life. Around 30 percent of inner-city traffic is caused by searching for a parking space. Drivers produce around 1.3 to 1.5 kg of CO2 per search. The probability of finding a free parking space without a dedicated application on a personal smartphone (“app”) is around 30 percent. With parking space search functions, this increases to around 80 percent. The availability of parking spaces is displayed in a desired area per street segment and in some cities an app navigates the driver directly there. The use of mobile phone data to predict parking spaces can be extended to the entire mobile network area of one or different mobile phone providers.

The DE 10 2013 211 632 A1 relates to a method for providing parking information about free parking spaces, in which historical and current information about available, free parking spaces is determined for at least one street, with at least from the information determined a parking segment, which includes one or more streets, is determined and statistical parameters about free parking spaces are generated for each parking segment from the information determined. Furthermore, a modeling is created for each parking segment, in which the information determined for the parking segment is processed in order to determine a parking status of the respective parking segment as a probability distribution (Pi).

At the US 8,484,151 B1 Parking availability for a geographic area is predicted using a parking availability model. First, the geographic population density is predicted for the geographic area. The predicted geographic population density is then applied to the parking availability model to produce a prediction of parking availability for the geographic area. The parking availability model includes a function that relates predicted geographic population density to parking availability for a geographic area. The projected parking availability for the geographic area is stored in a computer-readable storage medium. Parking availability predictions can be displayed as a layer within a map and can be created for a specifically requested geographic area or a general geographic area.

At the US 2017/191 849 A1 Parking-related statistics and real-time data as well as data received from routing data providers are normalized as normalized data. Using the normalized data, a computer calculates a probability of parking availability in specific geographic segments for a selected destination, routes to available parking spaces, the probability, and an estimated parking time (ETP) = estimated time to drive around (ETDA) + estimated walking time (ETW ) using the normalized statistical data and the selected target. The calculated routes to available parking spaces are evaluated in combination with user preferences to evaluate the calculated routes to available parking spaces. The ordered routes to available parking spaces are initiated for display on a graphical user interface (GUI).

Available forecasting services can be divided into two classes: first, the deterministic forecasts and second, the probability-based forecasts.

The first class of deterministic forecasts includes all detections of dedicated parking spaces via directly assigned sensors or equivalent status detections. This type of recording is particularly used in managed parking spaces (e.g. parking garages). An occupancy image of the parking spaces under consideration can be generated directly and online via portal connections. These systems require significant investments.

The second class of probability-based forecasts is based on statistical evaluations and models. To do this, there must be a statistical totality of measured values that ensures significance and thus allows a probability interpretation. In addition to classic mathematical statistics, so-called “machine learning” tools are also used, i.e. mathematical algorithms that can derive rules or classifiers from large data sets. Statistical methods are used in particular when direct refinancing of sensor equipment is not possible, for example in free parking on the side of the road. A possible static method is, for example, the generation of daily cycles, whereby the availability of parking space is recorded at suitable time intervals (e.g. every 10 minutes, every quarter of an hour, every hour) over a longer period of time for a twenty-four-hour day, based on a local unit (e.g. street segment, city district). The prediction then consists of the assumption that if there was a certain probability of an available/free parking space at a certain time of day at a certain location in the past with sufficient confidence, there will also be a free parking space at that time of day on another day in the future can be found.

In machine learning, for example, there is a data set that assigns a probability interpretation to certain situations. There are also data sources that have an inherent connection to the parking space occupancy situation (e.g. track data from vehicle movements, camera data, calendar data, ultrasonic sensors, selective ground sensors, etc.). The machine learning tools now create a mapping of this data to the verified situation, the so-called training. For use at execution time, the trained model is applied to the current data situation and a forecast is generated.

The following properties are important for machine learning in the present context. The way in which machine learning classifies input data always remains opaque; a direct interpretation is impossible. It follows that explicitly available knowledge is always of higher quality than algorithmic-statistical conclusions. Furthermore, machine learning methods have difficulty considering data on different time scales; In the present context, this concerns the comparison of validated daily cycles (see above) and short-term phenomena (e.g. 15 min). If the population is not chosen correctly, false precisions are created on the short-term scale.

The object of the underlying invention is to provide a method for forecasting available parking spaces that is particularly efficient or has low system costs, whereby the accuracy of the forecast for available parking spaces is as high as possible and is determined quickly with as little computing effort as possible. It is also an object of the invention to provide an arrangement that can carry out the method.

The present invention solves this problem by a method for predicting available parking spaces according to claim 1.

During the traffic structure analysis, the cells that have a comparatively high statistical significance are selected from the entirety of the cells in the cell grid. Based on the significant cells, a forecast for the other cells of the cell grid can be made with sufficient statistical certainty. For example, it has been shown that for a city that has a cell grid with 250 cells, only 9 significant cells are sufficient to create a parking probability forecast for all cells of the cell grid with sufficient accuracy. The calculation of the statistical parking situation is carried out on the basis of the real-time data set, with only the traffic data of the significant cells being used for the real-time data set. Since the number of significant cells in the cell grid is significantly lower than the number of all cells in the cell grid, significantly less traffic data needs to be registered and processed to determine the real-time data set. This advantageously leads to a significant increase in efficiency and thus to a reduction in system costs. For example, in a test, The underlying process reduces the system costs for providing traffic data in the form of mobile phone data by 90%. In particular, the system costs could be reduced because the method reduces the number of accesses required by an application on a mobile device to a server, the so-called “API calls”.

The individual steps are carried out on the basis of different data sets, so that the traffic structure analysis can be carried out independently of the calculation of the statistical parking situation and/or the sensor parking situation. In this way, the available computer resources can be divided according to needs. For the traffic structure analysis, for example, a longer first period of time can be used as the basis for the static data set, without the computationally complex traffic structure analysis having a negative impact on the computing time for calculating the statistical parking situation or the sensor parking situation. In contrast, the real-time data set is used to calculate the statistical parking situation, which contains significantly less information because, on the one hand, it only contains the traffic data from the significant cells and the second time period can be chosen to be shorter. Advantageously, less computing power and/or computing time is required, so that the statistical parking situation can be determined with sufficient accuracy based on a real-time data set obtained shortly before the forecast time.

For the probability forecast, the statistical parking situation determined based on the real-time data set is used in combination with the sensor parking situation. The sensor parking situation is determined using a sensor data set that is based on sensor data from the cells of the cell grid. The sensor data set is created independently of the other data sets and represents a direct measurement of the parking situation. The sensor data has the advantage that it can be evaluated immediately and does not require any complex statistical calculation, so that the sensor data set is relatively current, i.e. available shortly before the forecast time. By combining the statistical Parking situation with the sensor parking situation can advantageously achieve a higher accuracy in the forecast of available parking spaces. Preferably, a statistical parking probability is determined for at least one cell of the cell grid based on the significant cells based on the real-time data set. The statistical parking probability corresponds to a statistically significant probability of the existence of free parking spaces. In addition, a sensor parking probability is preferably determined for at least one cell of the cell grid based on the sensor data set. These values indicate the probability of a free/available parking space. Preferably, the respective statistical parking space probability and the respective sensor parking space probability are determined for each cell and used for the forecast.

A preferred embodiment of the invention is characterized in that during the traffic structure analysis, the cells of the cell grid are identified based on the static data set. The size, shape and/or position of the cells is particularly preferably determined based on the vehicle density, road density and/or parking space density. For example, the reference area of the static data set is divided into a large number of cells of a cell grid, whereby the cells can vary in size, shape and/or position. In particular, the cells are defined in such a way that they cover the reference area of the static data set as completely as possible without overlaps. The identification of the cells of the cell grid occurs before the selection of the significant cells. This advantageously increases the accuracy and efficiency of the forecast because the cells can already be defined in such a way that certain cells have a higher statistical significance. The cells of the cell grid preferably have a diameter of 50 to 5000 meters, particularly preferably 150 to 2500 meters.

According to the invention, the significant cells are selected based on a principal component analysis of the static data set. This corresponds a dimension reduction. Principal component analysis is a well-known statistical method for structuring large data sets by approximating a large number of statistical variables using a smaller number of linear combinations that are as meaningful as possible (the “principal components”). The advantage of selecting the significant cells using a principal component analysis is that well-known and proven algorithms can be used to evaluate the static data set.

• Fahrzeuge können (in einem statistisch relevanten Sinn; Unfälle oder Diebstähle fallen hier nicht ins Gewicht) weder erzeugt noch vernichtet werden. Dies bedeutet, dass bei geeigneter Wahl von Bezugsgebiet und Zeitintervall die Gesamtbilanz der Fahrzeuge Null ist (also alle Pendler im Mittel zu ihren Wohnungen zurückkehren oder alle Büroarbeitsplätze im Mittel abends verlassen werden).
• Fahrzeug, die aus einem Betrachtungsgebiet herausfahren, müssen in ein benachbartes Betrachtungsgebiet einfahren.
• Die Veränderung von Verkehrsstärken ist in einem statistischen Sinne stetig, Sprungverhalten von entsprechenden mathematischen Funktionen treten in der Realität nicht auf.
• Die Transportkapazität eines Straßensegments hat eine obere Grenze, die beispielsweise durch die geltende Höchstgeschwindigkeit, eine mittlere Fahrzeuglänge und den mittleren Abstand zum vorausfahrenden Fahrzeug gegeben ist.
• Der am Straßenrand verfügbare freie Parkraum (sofern nicht explizit bekannt oder über Verkehrsregelung eingeschränkt) hat eine berechenbare obere Grenze.

In a further preferred embodiment of the invention, explicit framework conditions or explicit model knowledge are linked to a statistical evaluation for traffic structure analysis. Examples of explicit model knowledge, or explicit framework conditions, in the subject area of urban traffic flows under consideration are:

• Vehicles can neither be created nor destroyed (in a statistically relevant sense; accidents or thefts are irrelevant here). This means that with a suitable choice of reference area and time interval, the overall balance of the vehicles is zero (i.e. all commuters return to their homes on average or all office workplaces leave on average in the evening).
• Vehicles leaving one observation area must enter an adjacent observation area.
• The change in traffic volumes is constant in a statistical sense; jump behavior of corresponding mathematical functions does not occur in reality.
• The transport capacity of a road segment has an upper limit, which is given, for example, by the applicable maximum speed, an average vehicle length and the average distance to the vehicle in front.
• The free parking space available on the side of the road (unless explicitly known or restricted by traffic regulations) has a calculable upper limit.

By using explicit frameworks, the identification of cells and the selection of significant cells can be localized Conditions of the cells are taken into account so that the accuracy of the forecast is increased.

A further embodiment of the invention is characterized in that the static data set and the real-time data set are based on traffic data that is obtained from mobile phone data. The mobile phone data is based on a total of mobile devices, in particular on a statistically relevant sample. The localization of mobile devices, which is carried out either through explicit GPS positioning or via registration in cell phones, enables a very precise picture of the mobility patterns in an area under consideration. The high event density of mobile radio-based data sets is preferably used to generate a particularly accurate image of the traffic situation. In particular, tracking paths of the terminal devices can be generated based on mobile phone data, and these paths can be analyzed using suitable statistical filters. Road users are particularly preferably identified based on their movement characteristics. Vehicles looking for parking spaces can also be identified based on their acceleration and braking behavior. Advantageously, the high event density of mobile phone data increases the accuracy of the statistical calculation. The use of mobile phone data to predict parking spaces can be extended to the entire mobile network area of one or different mobile phone providers. The static data set and the real-time data set can each be based on multiple data sets. In particular, mobile phone data for the data sets can come from different mobile phone providers.

A further embodiment of the invention is characterized in that the static data record is available off-line, ie is based on traffic data from a previous first period of time, which is already stored on a storage medium and is not recorded immediately before the forecast time. The statistical data set particularly preferably comprises a reference area which is completely covered by the cells of the cell grid. Advantageously, existing traffic data can be used to carry out the traffic structure analysis - a lengthy one Recording of traffic data is no longer necessary. The invention thus allows a combination of structural off-line analysis with real-time data.

A further embodiment of the invention is characterized in that the sensor data set is created on the basis of sensors that register events relevant to parking spaces, in particular an occupancy image is generated. The sensors are, for example, ground sensors, vehicle-based sensors, distance meters, cameras and/or infrastructure sensors. Thanks to the cost-effective and mass connection of sensor infrastructures, more and more areas of public life can be analyzed more comprehensively, including the area of vehicle-based mobility. Sensor data can be divided into stationary units that record parking space occupancy at one location and mobile sensors (e.g. installed in motor vehicles) that record parking space occupancy on the side of the road. For example, the sensor data set is created by a third party and made available for further use.

A further embodiment of the invention is characterized in that the time periods on which the data sets are based are different, with the first time period of the static data set being before the second time period of the real-time data set and the third time period of the sensor data set. Preferably, the second period of the real-time data set also lies before the third period of the sensor data set. The time periods preferably have different temporal dimensions, with the first time period of the static data set in particular being recorded years before the forecast time and taking years of traffic data into account. The first period preferably begins one to two years before the forecast date and covers a period of several months, ideally a year, in order to statistically record seasonal fluctuations in traffic. The second period of the real-time data set preferably begins 1.5 to 5 hours before the forecast time, particularly preferably two to three hours, and covers a period of at least one hour in order to obtain a statistically validated statement about the traffic balances in the cells under consideration. The third period of the sensor data set preferably begins 1 to 15 minutes before the forecast time, particularly preferably 5 to 10 minutes, and collects the current image of the sensor data for the traffic area under consideration, or reference area. It has been found that these values can advantageously be used to achieve the most precise forecast possible for the time periods.

A further embodiment of the invention is characterized in that a suitable mathematical combination of the statistical parking space probability and the sensor parking space probability is formed in the probability forecast. Preferably, the forecast of available parking spaces for a cell is the suitably weighted average of the statistical parking space probability and the sensor parking space probability. Advantageously, a correction/adjustment of the statistical calculation using actually available sensor data is achieved with little computational effort.

A further embodiment of the invention is characterized in that the method has a machine learning function in which a specific statistical parking situation is assigned to a specific sensor parking situation, in particular wherein the traffic structure analysis, the calculation and/or the probability forecast are adjusted/corrected. Advantageously, the accuracy of the forecast is improved by a machine learning function. In an exemplary embodiment of the invention, the machine learning function includes the mapping of a vector of measured sensor data at a few explicit locations at a time to the probable parking space occupancy situation at the same time for the entire traffic area under consideration. This image then acts as a further, now mathematically generated data set for the previously described data fusion.

A further embodiment of the invention is characterized in that a correction factor is used in the probability forecast if there is a systematic difference between the statistical parking space probability and the sensor parking space probability.

For example, when averaging, the statistical parking probability is multiplied by a correction factor if the statistical parking probability systematically deviates from the sensor parking probability. The correction factor is preferably variable and is adjusted depending on the difference between the probabilities.

The object on which the invention is based is also achieved by an arrangement for forecasting available parking spaces according to claim 11.

The arrangement is designed to carry out the aforementioned method. For this purpose, the arrangement has a storage unit for the static data set, a real-time processing unit for the real-time data set and a sensor communication unit for the sensor data set. The computer unit is intended to carry out the procedural steps that are carried out on the basis of the data sets. The real-time processing unit is used to evaluate the mobile radio data of the significant cells from the second period in order to determine the real-time data set. The sensor communication unit receives the sensor data from the sensors that measure parking-related events, for example floor sensors, vehicle-based sensors, distance meters, cameras and/or infrastructure sensors. The sensor data set is determined from the sensor data using the sensor communication unit. The computing unit determines the statistical parking situation, in particular the statistical parking probability, based on the data records, and the sensor parking situation, in particular the sensor parking probability. The units of the arrangement can be arranged spatially separately from one another or can form a spatial unit. Preferably, the computing unit is a server on which the method steps are carried out and with which the other units communicate.

A computer program product is also provided which, when the program is executed by an arrangement, causes it to carry out the steps of the previously described method. The computer program product controls the arrangement with the storage unit, the real-time data processing unit, the sensor communication unit and the computer unit in such a way that it carries out the traffic structure analysis, the calculation of the statistical parking situation and the sensor parking situation as well as the probability forecast for at least one cell of the cell grid.

• Kommunikation des mobilen Endgerätes mit der Recheneinheit der zuvor beschriebenen Anordnung;
• Übermittlung von der Prognose an verfügbaren Parkplätzen für zumindest eine der Zellen des Zellengitters von der Recheneinheit zum mobilen Endgerät;
• Verwendung der Prognose an verfügbaren Parkplätzen für zumindest eine der Zellen des Zellengitters zur Wiedergabe einer Parkplatzsituation und/oder zur Navigation eines Nutzers des mobilen Endgerätes.

A method for predicting available parking spaces is also provided, which includes the following steps carried out using the computer unit and a mobile device:

• Communication of the mobile terminal with the computing unit of the previously described arrangement;
• Transmission of the forecast of available parking spaces for at least one of the cells of the cell grid from the computing unit to the mobile terminal;
• Use of the forecast of available parking spaces for at least one of the cells of the cell grid to display a parking situation and/or to navigate a user of the mobile device.

The method is based on a mobile terminal communicating with the arrangement described above, with the forecast for available parking spaces for at least one cell determined by the arrangement being sent to the mobile device is transmitted. The forecast is used by the mobile device to display the parking situation and/or navigation to an available/free parking space. For example, for the cells in which the mobile device is located and for neighboring cells, the respective forecast for available parking spaces is displayed in the form of a numerical value or a color scale. Using this information, the user of the mobile device can find the cells in which they are most likely to get a free parking space. However, the forecast can also be used by a navigation unit, for example, whereby the navigation unit automatically guides the user to the cell in which the user is most likely to get a free parking space. The mobile device can be, for example, a smartphone, a tablet, a navigation device or vehicle on-board electronics, for example a head-up display.

A further computer program product is also provided which, when the program is executed by a mobile device, causes it to carry out the steps of the method described immediately above.

Fig. 1a:

zeigt eine schematische Darstellung eines durch Zellen gebildeten Zellengitters,

Fig. 1b:

zeigt eine schematische Darstellung einer Zelle eines Zellengitters.

Fig. 2:

zeigt schematisch die von einer Recheneinheit durchgeführten Verfahrensschritte zur Prognose von verfügbaren Parkplätzen und die Datensätze auf denen diese basieren,

Fig. 3:

zeigt eine schematische Darstellung einer Anordnung mit der Recheneinheit, die dazu eingerichtet ist das in Fig. 2 dargestellte Verfahren auszuführen, und

Fig. 4:

zeigt schematisch die von einem mobilen Endgerät durchgeführten Verfahrensschritte zur Prognose von verfügbaren Parkplätzen.

For example, the computer program product is an app that is executed on the mobile device and causes the mobile device to communicate with the computing unit of the arrangement, with the forecast for available parking spaces being transmitted to the mobile device. The transmitted forecast is used by the app to reflect the parking situation and/or to navigate to an available/free parking space. Preferred exemplary embodiments of the present invention are explained below with reference to the figures:

Fig. 1a:

shows a schematic representation of a cell grid formed by cells,

Fig. 1b:

shows a schematic representation of a cell of a cell grid.

Fig. 2:

shows schematically the process steps carried out by a computing unit to predict available parking spaces and the data sets on which these are based,

Fig. 3:

shows a schematic representation of an arrangement with the computing unit that is set up to do this Fig. 2 carry out the procedures presented, and

Fig. 4:

shows schematically the procedural steps carried out by a mobile device to predict available parking spaces.

Features of the present invention are explained below using preferred embodiments. The present disclosure is not limited to the specifically mentioned combinations of features. Rather, the features mentioned here can be combined in any way to form embodiments according to the invention, unless this is expressly excluded below.

In Fig. 1a Cells 1 of a cell grid 2 are shown. A specific reference area is covered by the cells 1 of the cell grid 2, which can be, for example, a city, a country or a traffic area, however defined. The cells 1 can have a different size or shape. The honeycomb shape shown is just an example. Preferably, the cells 1 of the cell grid 2 completely cover the relevant reference area, in particular without any overlap of neighboring cells 1. As in Fig. 1b shown, a cell 1 represents a selected area of the reference area, in which, for example, parking spaces 4 and streets 5 are located. There are various road users in cells 1, for example vehicles 6a, 6b or pedestrians 6c. Using mobile phone data from road users 6a, 6b, 6c, their position can be determined depending on time either through explicit GPS positioning or by booking in mobile phone cells. For example, lane formation is possible for the road users 6a, 6b, 6c, with a path of the corresponding terminal being available as a data record. The large number of mobile devices provides a precise picture of the mobility pattern in the reference area. Based on the mobile phone data of road users 6a, 6b, 6c as traffic data, a probability-based forecast for the availability of parking spaces can be created. Ideally, only mobile phone data is used for the parking forecast used by motor vehicles or mobile devices located in vehicles. The data set is previously divided/split into the different types of road users.

An embodiment of the method according to the invention for predicting available parking spaces is shown in Fig. 2 shown schematically. The method is based on a traffic structure analysis 11, a calculation 13 and a probability forecast 15. These method steps are carried out on the basis of different data sets 10, 12, 14. The time axis marked t shows the chronological sequence of the implementation of the individual method steps 10, 12, 14, with the probability forecast 15 taking place at the forecast time TP. The timeline is not to scale.

The traffic structure analysis 11 is carried out on the basis of a static data set 10. The static data record 10 is based on traffic data from a first period T1, with traffic data from all cells 1 of the cell grid 2 being used as a basis. The first period T1 is long before the forecast time TP and takes into account traffic data that has been recorded over a longer period of time. The first period T1 preferably begins one to two years before the forecast time and covers a period of several months, ideally one year. The traffic structure analysis 11 is thus created on the basis of a data set that has a high level of statistical significance. In this way, seasonal fluctuations in traffic are recorded statistically.

In the traffic structure analysis 11, significant cells 3 are selected from the cells 1 of the cell grid 2 based on the static data set 10. The significant cells 3 have a comparatively high statistical significance, so that a statistical forecast for other cells 1 of the cell grid 2 can be created with sufficient statistical certainty on the basis of the significant cells 3. The significant cells 3 are selected using a principal component analysis of the static data set 10.

In a preferred embodiment of the invention, the cells 1 of the cell grid 2 are identified in the traffic structure analysis 11 on the basis of the static data record 10. The identification of the cells 1 takes place before the selection of the significant cells 3. As part of the identification of the cells 1 of the cell grid 2, the size, shape and / or position of the cells 1 is determined, this being based on the vehicle density, road density and / or parking space density he follows. For example, cells 1 in metropolitan areas are smaller than cells 1 in rural areas. The cells 1 are preferably defined in such a way that they do not overlap and cover the reference area under consideration as completely as possible.

A real-time data set 12 and a sensor data set 14 are used for the calculation 13.

The real-time data set 12 is formed by traffic data from the significant cells 3, the traffic data from a second time period T2 being used. For the real-time data set 12, traffic data is used that is shortly before the forecast time TP and has been recorded over a comparatively short time. The second period T2 preferably begins two to three hours before the forecast time and covers a period of at least one hour. The mobile phone data of the road users who are in the significant cells 3 in the second period T2 act as traffic data. This allows the amount of data used for the real-time data set 12 to be significantly reduced, so that the efficiency of the forecasting process is improved and the system costs are reduced.

In the calculation 13, a statistical parking situation is determined for at least one of the cells 1 of the cell grid 2 based on the real-time data set 12. Preferably, a statistical parking probability is determined for the at least one cell 1, with the respective cell 1 being assigned a numerical value between 0 and 1, for example. The determination of the statistical parking situation of the respective cell 1 is accordingly carried out on the basis of the traffic data of the significant cells 3. A statistical method inverse to the main component analysis is used for the calculation. Because the real-time data set 12 is comparative is small, the statistical parking situation can be calculated quickly and with less computational effort.

The sensor data set 14 is based on sensor data from sensors that register parking-related events. When it comes to sensor data, a distinction is made between stationary sensors that register parking space occupancy in one location and mobile sensors, for example installed on motor vehicles, that record parking space occupancy on the side of the road. For the sensor data set 14, sensor data is used that lies before the forecast time TP. The third period T3 preferably begins 5 to 10 minutes before the forecast time and collects the current image of the sensor data situations for the traffic area under consideration. The sensor data is recorded for all or a selection of cells 1 of cell grid 2.

In the calculation 13, a sensor parking situation is calculated for at least one of the cells 1 of the cell grid 2 based on the sensor data set 14. Preferably, a sensor parking probability is determined for at least one cell 1, with the respective cell being assigned a numerical value between 0 and 1, for example. By using the current sensor data to calculate the sensor parking situation, the accuracy of the forecast for available parking spaces can be improved because, in addition to the statistical calculation, the actual situation is taken into account.

The probability forecast 15 for available parking spaces is carried out at the forecast time TP based on the statistical parking situation and the sensor parking situation. Preferably, in the probability forecast 15, a suitable mathematical combination of the statistical parking space probability and the sensor parking space probability is formed for at least one cell 1 of the cell grid 2. This is preferably a suitably weighted average. This allows the statistical calculation to be easily corrected using the actual parking situation recorded by the sensors.

In an embodiment of the method according to the invention, which is not shown, it has a machine learning function in which a specific statistical parking situation is assigned to a specific sensor parking situation. In particular, the machine learning function adapts or corrects the traffic structure analysis 11, the calculation 13 and/or the probability forecast 15. The machine learning function creates a mapping of the statistical parking situation to the sensor parking situation, the so-called training. For use at the time of forecasting, the trained model is applied to the current data situation.

A practical example of a machine learning function is that the probability prediction 15 uses a correction factor when there is a systematic difference between the statistical parking probability and the sensor parking probability. For example, if it is determined that for a cell 1 the statistical probability of parking is regularly twice as high as the sensor parking probability, the statistical probability for this cell 1 is multiplied by a factor of 0.5.

Fig. 3 shows a schematic representation of an arrangement 20 and a mobile terminal 30 for predicting available parking spaces. The arrangement 20 has a storage unit 21, which is designed to store the static data record 10 and make it available for further calculation, for example the traffic structure analysis 11. In addition, the arrangement 20 has a real-time data processing unit 22, which is designed to determine the real-time data set 12 from the traffic data of the significant cells 3. For this purpose, the real-time data processing unit 22 is able to read out and evaluate the mobile radio data of the road users 6a, 6b, 6c, which are located in the significant cells 3 in the second time period T2. When evaluating the mobile phone data, road users 6a, 6b, 6c in particular are identified based on their movement characteristics. For example, vehicles 6a looking for a parking space can be identified based on their acceleration and braking behavior. The arrangement 20 also has one Sensor communication unit 23, which is set up to determine the sensor data set 14 from sensor data. For this purpose, the sensor communication unit 23 establishes a connection with the sensors present in the reference area and creates a sensor data set 14 immediately before the forecast time TP. The traffic structure analysis 11, the calculation 13 and the probability forecast 15 are carried out by means of a computing unit 24. The computing unit 24 is designed to identify the cells 1 of the cell grid 2 from the static data record 10 and to select the significant cells 3 of the cell grid 2. In addition, the statistical parking situation is calculated with the computing unit 24 based on the real-time data set 12 and the sensor parking situation is calculated based on the sensor data set 14. The components of the arrangement 20 can be spatially separated or form a spatial unit. The arrangement 20 and/or the computing unit 24 is preferably a server.

The mobile terminal 30 is connected to the arrangement 20 via a network 25. The mobile terminal 30 is, for example, a smartphone, a navigation device or on-board electronics of a motor vehicle. The mobile terminal 30 is designed to communicate with the arrangement 20, in particular its computing unit 24, and to receive a forecast of available locations at least one of the cells 1 of the cell grid 2 from the arrangement 20. The forecast received from the arrangement 20 is used by the mobile terminal 30 to display the parking situation and/or to carry out navigation for the user of the mobile terminal 30.

In Fig. 4 A method is shown in which the prediction of available parking spaces is carried out using the arrangement 20 and the mobile terminal 30. First, communication 40 is established or carried out between the mobile terminal 30 and the arrangement 20. During communication 40, the forecast of available parking spaces for at least one of the cells 1 of the cell grid 2 is transmitted 41 from the arrangement 20 to the mobile terminal 30. Using 42 the forecast of available parking spaces, the parking situation is determined using the mobile Terminal 30 played and / or used by means of the mobile terminal 30 for navigation of a user. For example, the current position is recorded by the mobile terminal 30 and the parking situation of the corresponding cell 1 and/or the neighboring cells 1 is represented by numerical values or color coding. This in Fig. 4 The method shown can be carried out by a program product, or an app, running on the mobile terminal 30.

Those looking for parking spaces, cities and the environment benefit equally from the procedures, the arrangement and the program product. Drivers reach their destination faster and more relaxed. Municipalities optimize traffic and thus increase the attractiveness of city centers for residents, trade and tourism. In addition, the air is less polluted with pollutants.

List of reference symbols:

1

cell

2

Cell grid

3

significant cell

4

parking spot

5

Street

6a

Road users/motor vehicles (looking for a parking space)

6b

Road users/motor vehicles (not looking for a parking space)

6c

Road users/pedestrians

10

static data set

11

Traffic structure analysis

12

Real-time data set

13

calculation

14

Sensor data set

15

Probability forecast

20

arrangement

21

Storage unit

22

Real-time data processing unit

23

Sensor communication unit

24

Computing unit

25

network

30

mobile device

40

communication

41

Transmission of the forecast

42

Usage (Playback/Navigation)

T1

first period

T2

second period

T3

third period

TP

forecast time

Claims (11)

1. A method for forecasting available parking spaces, comprising the following steps performed with the aid of a computing unit (24):

   - a traffic structure analysis (11), in which significant cells (3) are selected from cells (1) of a cell matrix (2) based on a static dataset (10) formed by traffic data from a first time period (T1), wherein a forecast with sufficient statistical probability can be prepared for the other cells (1) of the cell matrix (2) based on the significant cells (3), wherein the selection of the significant cells (3) is carried out based on a principal component analysis of the static dataset (10) or another mathematical dimensionality reduction method;

   - a calculation (13) in which a statistical parking space situation, in particular a statistical parking space probability is calculated for at least one of the cells (1) of the cell matrix (2) based on a realtime dataset (12) formed by traffic data of the significant cells (3) from a second time period (T2), and

   - in which a sensor parking space situation, in particular a sensor parking space probability is calculated for at least the one of the cells (1) of the cell matrix (2) based on a sensor dataset (14) formed by sensor data of the cells (1) from a third time period (T3); and

   - a probability forecast (15) in which a forecast of available parking spaces is carried out for at least the one of the cells (1) of the cell matrix (2) based on the statistical parking space situation, in particular the statistical parking space probability, and the sensor parking space situation, in particular the sensor parking space probability.
2. The method for forecasting available parking spaces according to Claim 1, characterized in that in the traffic structure analysis (11) the cells (1) of the cell matrix (2) are identified based on the static dataset (10), in particular wherein the size, shape and/or position of the cells (1) are determined with on the basis of the vehicle density, street density and/or parking space density.
3. The method for forecasting available parking spaces according to any one of the preceding claims, characterized in that in the traffic structure analysis (11) explicit frame conditions are linked to a statistical evaluation, in particular wherein the frame conditions are for example that with suitable selection of reference domain and time interval the overall balance of vehicles is equal to zero, that the change in the traffic continues constantly, that the transport capacity of a road segment has an upper limit, and/or that the available parking area of a road segment has an upper limit.
4. The method for forecasting available parking spaces according to any one of the preceding claims, characterized in that the static dataset (10) and the realtime dataset (12) are based on traffic data obtained from mobile radio data, in particular wherein road users are identified on the basis of their movement characteristics, and in particular wherein vehicles searching for parking spaces are identified on the basis of their acceleration and braking behaviour.
5. The method for forecasting available parking spaces according to any one of the preceding claims, characterized in that the static dataset (10) is available off-line, in particular wherein the statistical dataset (10) comprises a reference domain that is entirely covered by the cells (1) of the cell matrix (2).
6. The method for forecasting available parking spaces according to any one of the preceding claims, characterized in that the sensor dataset (14) is prepared on the basis of sensors that record events of significance for parking spaces, for example the sensors are ground sensors, vehicle-based sensors, distance meters, cameras and/or infrastructure sensors.
7. The method for forecasting available parking spaces according to any one of the preceding claims, characterized in that the time periods (T1, T2, T3) are different, wherein the first time period (T1) occurs before the second time period (T2) and the third time period (T3), in particular wherein the second time period (T2) occurs before the third time period (T3).
8. The method for forecasting available parking spaces according to any one of the preceding claims, characterized in that in the probability forecast (15) a mean value is generated from the statistical parking space probability and the sensor parking space probability.
9. The method for forecasting available parking spaces according to any one of the preceding claims, characterized in that the method includes a machine learning function, in which a certain statistical parking space situation is assigned to a certain sensor parking space situation, in particular in order to adapt and/or correct the traffic structure analysis (11), the calculation (13) and/or the probability forecast (15).
10. The method for forecasting available parking spaces according to Claims 9 and 10, characterized in that in the probability forecast (15) a correction factor is used if a systematic difference exists between the statistical parking space probability and the sensor parking space probability.
11. An apparatus (20) for forecasting available parking spaces, comprising

    - a storage unit (21) which is designed to store a static dataset (10) formed by traffic data for a first time period (T1);

    - a realtime-data processing unit (22) which is designed to determine a realtime dataset (12) through traffic data of significant cells (3) from a second time period (T2);

    - a sensor communication unit (23) which is designed to determine a sensor dataset (14) from sensor data from a third time period (T3); and a

    - computing unit (24) which is designed to carry out the method steps of Claim 1.

Images