WO2020035383A1

WO2020035383A1 - Method and system of recommending a place to park

Info

Publication number: WO2020035383A1
Application number: PCT/EP2019/071296
Authority: WO
Inventors: Wei Ming Chia; Bin Peng; Xue Liu; Jordan IVANCHEV; Vaisagh VISWANATHAN; Suraj Nair; Michael POPOW
Original assignee: Continental Automotive Gmbh
Priority date: 2018-08-13
Filing date: 2019-08-08
Publication date: 2020-02-20
Also published as: GB201813149D0; GB2576312A; SG11202012037SA

Abstract

The invention relates to a method of recommending at least one available car park amongst multiple car parks within a geographical area in response to a query of a current user received by a user device, the method being executed in a system remote from the user device. The method comprises retrieving query data (102) and past decision data (104).The method comprises: analysing car park data (106) of at least one car park within the geographical area, using a first model (110), in order to predict car park occupancy for a specified time window in the future; analysing the query data (102) and the car park occupancy predicted by the first model, using a second model (112), in order to refine the prediction of the car park occupancy from the first model (110); analysing the past decision data (104) and the refined prediction (108) of the car park occupancy from the second model,using a third model (114), in order to recommend at least one available car park within the geographical area. The invention further relates to a computer program product that when in use can perform the disclosed method. The invention also relates to a system (100) of recommending at least one available car park amongst multiple car parks within a geographical area in response to a query of a current user received by a user device remote from the system.

Description

Method and System of Recommending a Place to Park

FIELD OF INVENTION

This invention relates to a method and system of recommending at least one available car park amongst multiple car parks within a geographical area in response to a user guery.

BACKGROUND OF INVENTION

In urban areas, the search for an available parking lot may cause stress to drivers due to factors like dense traffic and one-way roads. Thus, technology has emerged to provide drivers with knowledge of an available parking lot.

For example, it is known to use sensors to provide information on real-time car park occupancy. It is also known to use crowd-sourced locations to determine where available parking spots are located. Other known methods include guiding drivers to parking locations where the drivers are likely to find parking spots. However, these methods direct all drivers to the same available car park, thereby concentrating traffic density to the same area, potentially causing congestion in that area. Furthermore, car park occupancy may not be accurately predicted at the time the parking spot is needed.

There is therefore a need to provide an alternative method for locating available parking lots while optimizing traffic flow for other users of the road.

SUMMARY

It is therefore an object to provide a method and system to address the problems discussed above. Particularly, it is an object to provide a method of recommending suitable parking spaces that minimizes the overall congestion in an area. It is an object to provide recommendations of parking spaces that minimizes the overall congestion in an area. It is a further object to use parking recommendations as a steering tool to influence drivers to achieve an optimum traffic density, minimizing congestion for every traffic participant.

To accomplish these and other objects, there is provided a method of recommending at least one available car park amongst multiple car parks within a geographical area in response to a query of a current user received by a user device, wherein the method is executed in a system remote from the user device, the method comprising: retrieving a query data feed comprising the query of the current user and a query of at least one other current user, each current query comprising: an intended parking location in a specified time window in the future defined by each current user, wherein the intended parking location is within the geographical area; retrieving a past decision data feed comprising a final parked location of at least one past user; analysing car park data of at least one car park within the geographical area, using a first model implemented by the processor, in order to predict car park occupancy for the specified time window in the future, wherein the car park data comprises current and historical car park occupancy retrieved from at least one public database; analysing the query data feed and the car park occupancy predicted by the first model, using a second model implemented by the processor, in order to refine the prediction of the car park occupancy from the first model; analysing the past decision data feed and the refined prediction of the car park occupancy from the second model, using a third model implemented by the processor, in order to recommend at least one available car park within the geographical area. The disclosed method advantageously takes into account historical and current data, and both publicly available data as well as user data, to predict future car park occupancy. The query data feed provides information about current users who are looking for parking spots, which is information that publicly available car park data may not be able to catch. The current batch of drivers looking for a car park or parking spot in the same area is considered in the prediction. Accordingly, a more accurate picture of the car park occupancy at the future time point may be provided by the disclosed method.

Additionally, because users may not finally choose to park at the recommended car park(s), the disclosed method advantageously takes into account the final parking decision of past users. Not only is the knowledge of actual historical car park occupancy, e.g. from the public databases, used as an input into the disclosed method, the effect of actions of past users may be used as input. Past users' actions may be used to determine how current users may select a car park. This information may further help to improve the prediction of car park occupancy at the future time point. The more accurate the car park occupancy prediction, the better the recommendation of suitable available car park(s) .

The third model may advantageously be selected to balance traffic or distribute traffic substantially uniformly within the geographical area to minimize traffic congestion. Thus, the current users may also advantageously be substantially uniformly distributed within the geographical area.

There is therefore provided, in an aspect, a method of generating an improved recommendation of at least one available car park amongst multiple car parks within a geographical area in response to a query of a current user received by a user device, wherein the method is executed in a system remote from the user device, the method comprising: retrieving a guery data feed comprising the guery of the current user and a guery of at least one other current user, each current guery comprising: an intended parking location in a specified time window in the future defined by each current user, wherein the intended parking location is within the geographical area; retrieving a past decision data feed comprising a final parked location of at least one past user; analysing car park data of at least one car park within the geographical area, using a first model implemented by the processor, in order to predict car park occupancy for the specified time window in the future, wherein the car park data comprises current and historical car park occupancy retrieved from at least one public database; analysing the guery data feed and the car park occupancy predicted by the first model, using a second model implemented by the processor, in order to refine the prediction of the car park occupancy from the first model; analysing the past decision data feed and the refined prediction of the car park occupancy from the second model, using a third model implemented by the processor, in order to generate a recommendation of at least one available car park within the geographical area that causes the current user and the at least one other current user to be substantially uniformly distributed within the geographical area.

The disclosure provides the possibility for achieving cooperative optimum load balancing of traffic using parking lot recommendation. With the objective of load balancing of the traffic system, the disclosed method may take as input the user gueries which define which parking lots they want to go to, the decisions made by users in previous guery cycles and the predicted state of the car parks. The disclosed method may make use of available historical car park data to train models, such as machine learning models, and the trained models may be used together with real-time user guery data and parking decisions to provide prediction or forecasting of the state of car parks over the specified time window. As a response to each query, the disclosed method may generate a score card for every user which contains a list of recommended car parks such that even if only a fraction of the users accepts the recommendation, an optimal distribution of traffic on the roads can be achieved.

In another aspect, there is provided a computer program product residing on a non-transitory computer readable storage medium of a system, the storage medium having machine-readable instructions stored thereon which, when executed by the computer, cause the system to perform the disclosed method.

There is also provided a system of recommending at least one available car park amongst multiple car parks within a geographical area in response to a query of a current user received by a user device remote from the system, the system comprising: a database server configured to receive and store: car park data of at least one car park within the geographical area, the car park data comprising current and historical car park occupancy retrieved from at least one public database; query data comprising the query of the current user and a query of at least one other current user, each current query comprising: an intended parking location in a specified time window in the future defined by each current user, wherein the intended parking location is within the geographical area; past decision data comprising a final parked location of at least one past user; an analytical server configured to analyse the car park data of each car park using a first model, in order to predict car park occupancy for the specified time window in the future, the analytical server further configured to analyse the query data and the car park occupancy predicted by the first model, using a second model, in order to refine the prediction of the car park occupancy from the first model, the analytical server further configured to analyse the past decision data and the refined prediction of the car park occupancy from the second model, using a third model, in order to recommend at least one available car park within the geographical area.

In an aspect, there is provided a system of generating an improved recommendation of at least one available car park amongst multiple car parks within a geographical area in response to a guery of a current user received by a user device remote from the system, the system comprising: a database server configured to receive and store: car park data of at least one car park within the geographical area, the car park data comprising current and historical car park occupancy retrieved from at least one public database; guery data comprising the guery of the current user and a guery of at least one other current user, each current guery comprising: an intended parking location in a specified time window in the future defined by each current user, wherein the intended parking location is within the geographical area; past decision data comprising a final parked location of at least one past user; an analytical server configured to analyse the car park data of each car park using a first model, in order to predict car park occupancy for the specified time window in the future, the analytical server further configured to analyse the guery data and the car park occupancy predicted by the first model, using a second model, in order to refine the prediction of the car park occupancy from the first model, the analytical server further configured to analyse the past decision data and the refined prediction of the car park occupancy from the second model, using a third model, in order to generate a recommendation of at least one available car park within the geographical area that causes the current user and the at least one other current user to be substantially uniformly distributed within the geographical area. As the present disclosure aids in generating recommendations that substantially uniformly distribute users within a geographical area, the users may advantageously experience less congestion and consequently save more time when looking for available parking lots amongst a plurality of car parks within a geographical area.

DETAILED DESCRIPTION

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. The detailed description of this invention will be provided for the purpose of explaining the principles of the invention and its practical application, thereby enabling person skilled in the art to understand the invention for various exemplary embodiments and with various modifications as are suited to the particular use contemplated. The detailed description is not intended to be exhaustive or to limit the invention to the precise embodiments disclosed. Modifications and equivalents will be apparent to practitioners skilled in this art and are encompassed within the spirit and scope of the appended claims.

Fig. 1 shows an illustration of the system architecture according to an embodiment of this invention. Specifically, the figure illustrates the different components and sub-components of the disclosed system 100 which support the disclosed method. The following description will make reference to the reference numerals specified in the figure.

In an embodiment, the method disclosed is a method of recommending at least one available car park amongst multiple car parks within a geographical area in response to a query of a current user received by a user device. The method is executed in a system 100 remote from the user device. The method comprises retrieving a query data feed 102 comprising the query of the current user and a query of at least one other current user, each current query comprising: an intended parking location in a specified time window in the future defined by each current user, wherein the intended parking location is within the geographical area. The method comprises retrieving a past decision data feed 104 comprising a final parked location of at least one past user. The method comprises analysing car park data 106 of at least one car park within the geographical area, using a first model 110 implemented by the processor, in order to predict car park occupancy for the specified time window in the future, wherein the car park data 106 comprises current and historical car park occupancy retrieved from at least one public database. The method comprises analysing the query data feed 102 and the car park occupancy predicted by the first model, using a second model 112 implemented by the processor, in order to refine the prediction of the car park occupancy from the first model. The method comprises analysing the past decision data feed 104 and the refined prediction 108 of the car park occupancy from the second model, using a third model 114 implemented by the processor, in order to recommend at least one available car park within the geographical area.

In another embodiment, the method disclosed is a method of generating an improved recommendation of at least one available car park amongst multiple car parks within the geographical area in response to the current user query received by the current user's device. The method comprises, inter alia, analysing the past decision data feed 104 and the refined prediction 108 of the car park occupancy from the second model, using the third model 114 implemented by the processor, in order to generate a recommendation of at least one available car park within the geographical area that causes the current user and the at least one other current user to be substantially uniformly distributed within the geographical area.

In an embodiment, the disclosed system 100 is a system of recommending at least one available car park amongst multiple car parks within a geographical area in response to a query of a current user received by a user device remote from the system 100. The system 100 comprises: a database server 120 configured to receive and store: 1. car park data 106 of at least one car park within the geographical area, the car park data 106 comprising current and historical car park occupancy retrieved from at least one public database; 2. query data 102 comprising the query of the current user and a query of at least one other current user, each current query comprising: an intended parking location in a specified time window in the future defined by each current user, wherein the intended parking location is within the geographical area; 3. past decision data 104 comprising a final parked location of at least one past user. The system 100 comprises an analytical server (130) configured to analyse the car park data 106 of each car park using a first model 110, in order to predict car park occupancy for the specified time window in the future. The analytical server 130 is further configured to analyse the query data 102 and the car park occupancy predicted by the first model, using a second model 112, in order to refine the prediction of the car park occupancy from the first model. The analytical server 130 is also configured to analyse the past decision data 104 and the refined prediction 108 of the car park occupancy from the second model, using a third model 114, in order to recommend at least one available car park within the geographical area.

In another embodiment, the disclosed system 100 is a system of generating an improved recommendation of at least one available car park amongst multiple car parks within the geographical area in response to the current user query received by the current user's device. The analytical server 130 of the system 100 is configured to, inter alia, analyse the past decision data 104 and the refined prediction 108 of the car park occupancy from the second model, using the third model 114, in order to generate a recommendation of at least one available car park within the geographical area that causes the current user and the at least one other current user to be substantially uniformly distributed within the geographical area.

The term "car park" refers to a parking facility. A car park may be a group of parking lots, parking spots or parking spaces, for example along a stretch of road. A group of parking lots may stretch from a junction, e.g. a T-junction or a cross-junction, to another junction, or may be a segment thereof. The road segment may include one direction of traffic or both directions of traffic. The group of parking lots may be housed in a single storey or multi-storey building. The building may be a dedicated parking facility or shared with other purposes, e.g. commercial shops. A car park may be freely accessible to the public, or may be bounded by a barrier, e.g. a fence, to restrict access , or bounded by a barrier, e.g. kerb(s), to define a car park zone. The term "car park" is not limited to a parking facility for cars, but for any type of vehicle. The term "parking facility" is synonymous with the terms "parking garage" and "car park". Typically, a car park has an identification (ID), e.g. a car park name or identification number, given by the management or owner of the car park, or a state authority if the car park is stated-owned . The recommendation may refer to such car park ID. A user may refer to such car park ID in a query.

The current user, as referred to herein, is the user, e.g. a driver, querying the system 100 or using the method to find a suitable car park in a certain area and at a certain time. There may be other users querying the system 100 or using the method at the same time, but with different query parameters, e.g. different destination or different intended parking time. Users may be required to set up an account with the provider of the disclosed system and method so that user queries and decisions can be tracked and retrieved. Thus, as accounts are continually being created or deleted and as users need car park recommendations at different times, the group of past users may differ from the group of current users. In general, there may be N users and each of User 1 to User N may provide a current query to query data feed 102 and may contribute the final parked location from a past query to past decision data feed 104.

When a driver is searching for a car park, there may be a limit or a maximum distance from the intended destination that the driver is willing to park. Thus, the disclosed method recommends car parks within a geographical area with this limit taken into consideration. The geographical area may be defined by a predetermined distance from the current user's intended destination or intended parking location. The geographical area may be defined by the provider of the disclosed method or by a user of the method. For example, the user may define that only parking lots no further than a predetermined distance (e.g. five blocks or 500 meters) from the user's intended destination location are considered. The predetermined distance may be a radius .

The term "data feed" as used herein refers to a stream or collection of data that may be fed or delivered to a data collector 120, or may be fed directly to the models for analysis . Therefore, the user query data feed 102 may be a data stream of user queries from the current user and other current users. The past decision data feed 104 may be a data stream of the final parked locations of past users, past decisions and/or past user queries. Car park data 106 of each car park may be a data feed of car park information such as current and historical car park occupancy.

The data collector 120 may be a database server. The data collector 120 or database may be located in a location remote from the user device. The data collector 120 may be located in one location or multiple locations, in one or more servers in one or more locations. The data collector 120 may be a computer system in itself and may store and control data feeds going in and out of the data collector 120.

The user device includes means of receiving user input, outputting car park recommendations, and may include a positioning device. The user device may be a mobile phone, a wearable device, a portable computer, a navigation system in a vehicle, etc. Score card 140 may be presented on the user device.

A user query 102 may include a user's input of an intended destination, for example an intended parking location. The intended parking location may be input as a car park ID. The geographical area may then be defined as the predetermined distance, e.g. a radius, around the intended destination. The user query 102 may include an input of when the user is expecting to reach the intended destination. The estimated time of arrival at the intended destination represents the specified time window in the future defined by the user, e.g. by user input or by input from an appropriate database, such as a digital map, upon identifying the source location of the query and the intended destination location. The user query 102 may include the source location of the query or the location of the user at the time of making the query. The user query 102 may include the time stamp when the query is entered. The source location and the time of the query may be input by the user or may be retrieved from the computer system or a digital map or from any other appropriate database. The user query 102 may include the search results of the recommended car parks of other current users .

The past decisions 104 may include the final parked location of past users. Each time a user makes a decision and decides to route to a particular car park, the car park chosen may be stored in a database, e.g. the car park ID. The past decisions 104 may include past user queries, and may include the time the past user parked the vehicle, the duration of parking, the source location when the past user made the query, the time when the past query was made, and the recommended car parks of past users. The current user may contribute to the past decision data feed 104, or may be a new user of the disclosed method and thus may not have any historical decisions or historical user queries.

Car park data 106 of each car park may include identity of the car park, the total capacity of the car park, and the number of available lots or car park occupancy at a specific time. The car park data 106 may include current car park occupancy and historical car park occupancy. The car park occupancy may be retrieved from a public database (not shown) . By "public database", it is meant that the data therein is retrievable by any person for a fee or for free.

The car park data collection process generally requires interfacing with disparate systems through possibly different protocols (e.g. HTTP, SOAP, etc.) and different output formats (e.g. XML, JSON, CSV, etc.) . The data collector 120 may periodically query the external, public databases in the disparate systems and imports the required data into a temporary database. The car park data 106 may be retrieved from the public database at a predetermined frequency. A querying frequency from the external databases may have to be determined to find a good balance between high-resolution information and managing load on the servers or the data collector 120. Due to various factors, queries often fail to return data or may even return erroneous information. When querying of the car park databases fails, the data collector 120 may be configured not to query again as it may be unnecessary and may likely be compensated for by the next query. In the temporary database, the car park data retrieved may be cleaned, filtered and categorized into individual car parks (Car Park 1, Car Park 2, ... Car Park N) for analysis by the models. The car park data retrieved may be parsed to the required format. The filtering of car park data is a process by which erroneous data may be dealt with. Due to errors that can happen in any part of the car park data collection process (sensor issues, communication issues, etc), it is possible that any of the databases, e.g. the public car park database or the data collector 120, stores inaccurate data. Inaccurate data is generally in the form of a sudden drop of available parking spaces in a very short period of time, generally followed by a sudden rise in available spaces. Such sudden spikes and dips may be dealt with by smoothing using a floating average over the last few values.

The temporary database may be separate from the data collector 120 or may be part of the data collector 120. The car park data in the temporary database may be fed into the data collector 120 for providing to the disclosed method. The car park data in the temporary database may be provided directly to the disclosed method. Alternatively, the data collector 120 may query, clean, filter and/or categorize the car park data for provision to the disclosed method.

The analytical server 130 may undertake the analysing of the data feeds. The analytical server 130 and the data collector 120 may be hosted on one server, or may be in separate servers in one location or multiple locations. The analytical server 130 may be configured to store and execute the analysis of the first model 110, the second model 112, and the third model 114.

The first model 110 analysing the car park data 106 may be a machine learning model for training the data to predict car park occupancy for the specified time window in the future. The machine learning model may include statistical and/or probabilistic modelling. The first model 110 may comprise an artificial neural network, making use of available historical data of car parks to train artificial neural network models which are thereafter used together with real time data of users and parking lot states to provide forecasting over a time window. The artificial neural network may include any suitable model, such as a deep neural network model or convolutional neural network model or a combination of models . It has been found that neural network models providing hidden nodes or weights, such as artificial neural networks, result in better prediction forecasting compared to neural networks with known weights, such as logistic regression. The car park data 106 fed into the first model 110 may include current and historical car park occupancy of one car park. The car park occupancy patterns may also be analysed to determine the duration a car park slot is occupied, to provide more specific occupancy information for the specified time window in the future. The first model 110 may be trained on each car park 1 ... N . As the car park data 106 may be retrieved from external databases at certain time intervals, the current car park occupancy may refer to the most recent car park occupancy retrieved. Furthermore, some external databases may not contain an extensive history of car park data. The accuracy of the future occupancy predicted by the first model for such car parks may therefore not be as high as for car parks with more extensive historical data. The first model 110 may analyse historical car park data 106 for each car park for fixed intervals in the past and/or for the full available history. For example, the first model 110 may analyse occupancy trends of a car park for the past hour, including the current occupancy trend, to predict future occupancy of the car park. The car park occupancy for each car park 1 ... N may be predicted for any specified time window in the future as defined by the user or estimated by a digital map. The duration of historical data used and the future time window of prediction may be different for each car park's model. Separate models may be used for each car park and for different durations of predictions. In general, the further into the future is the prediction interval, the larger the sample of historical data that will be needed as more data helps to improve the accuracy of the prediction model for longer forecasting periods. Alternatively, the car park occupancy for each car park 1 ... N may be predicted by the first model for fixed intervals in the future as defined by the computer system, for example every 5 minutes, every 15 minutes , or every hour. The second model 112 and/or third model 114 may then extrapolate or interpolate from the occupancy predicted by the first model 110 to arrive at the specified future time window defined by the user or estimated by the digital map. The output of the first model may therefore help to determine the car park occupancy patterns throughout the day, thereby aiding in predicting future occupancy patterns for the specified time window in the future from the current query.

The second model 112 analysing the query data feed 102 and the car park occupancy predicted by the first model may be a machine learning model for refining the prediction of the first model. The second model 112 may comprise a regression model. The user guery data feed 102 is used as input to the trained model to refine the prediction of the car park occupancy for the required future moment in time. In cases where the prediction interval is too far in the future for the chosen first and second models to adequately predict future occupancy, the second model 112 may be programmed to combine machine learning models, such as a combination of an artificial neural network and regression, to give a reasonable prediction estimate. The first and second models may be a combination of machine learning models, such as a combination of an artificial neural network and regression. The combination of first and second models may receive car park data 106, such as current and historical car park occupancies for a fixed interval in the past, e.g. for the past hour, to provide an adeguate prediction of future occupancy of the car park. In general, the output of the second model 112 or combination of first and second models is a refined prediction 108 of the state of each car park.

The third model 114 analysing the past decision data feed 104 and the refined prediction 108 of the car park occupancy from the second model may be configured to substantially uniformly distribute the current user and the at least one other current user within the geographical area. As mentioned above, past users' decisions may be used to estimate how current users may select a car park given a list of recommended car parks. Therefore, the third model 114 may be termed as an "adaptive load balancer". The query data feed 102 may be a further input into the third model 114 so that the raw current queries may be considered in distributing the car park load in the specified future time window. The current car park occupancy from the external database or current occupancy estimated from the historical car park data may be a further input into the third model 114. The car park data may be a further input into the third model 114. Therefore, the third model 114 may be a complex model that takes input from various models and/or data feeds in order to give recommendations to users in a way optimized for the particular goal, which is to balance the load of vehicles on the road network and thereby minimize traffic congestion. This disclosure therefore helps to improve traffic and the efficiency of car parks. The third model 114 may comprise different types of traffic system projections. A traffic system projection may take into account the state of historical and/or current traffic characteristics, e.g. speed, density, flow, etc., of a road network. The traffic characteristics may be detected by sensors that detect roadway utilization, e.g. speed sensors or traffic cameras, and stored in a database. Such database comprising traffic system characteristic data is typically administered by a state or government transportation authority. The third model 114 may comprise statistical models or traffic simulations. A suitable model may include a mesoscopic traffic simulation model which may determine the optimal balance. These projections predict future road usage and calculate what would be the optimal location for the user to be directed to, such that traffic congestion is minimized on a system level, thus minimizing travel time for all users as a consequence.

A recommendation that causes the current user and the at least one other current user to be substantially uniformly distributed within the geographical area, may be generated. The present disclosure advantageously provides improved technical solutions to generate recommendations of available parking lots that optimize the traffic density in a particular location. The generated recommendations are therefore more likely to be desired by the user since the user can avoid congestion and save time as a consequence. The use of the third model particularly generates improved parking lot recommendations that can minimize overall traffic congestion in an area and are therefore more likely to be desired by the user.

A web application presents the optimal location calculations from the analytical server as a choice or a list of car parks with scores, which may in turn be provided to the user via the user device. The web application, the analytical server 130 and the data collector 120 may be hosted on one server, or may be in separate servers in one location or multiple locations. The output of the disclosed method is used in a decision support system which responds to user queries about availability of parking lots through a score based model. A score card 140 listing the multiple car parks in order of recommendation may be provided to the user device, wherein the recommendation is based on car park occupancy and the distribution of the current user and the at least one other current user within the geographical area. The list of recommendations may be colour coded based on the order of recommendation, e.g. green for a car park (Car Park A) that is available and is in an area within the geographical area with the least traffic congestion, or red for a car park (Car Park C) that has few available lots and is in an area with high density of traffic. The score card 140 or list of recommendations indicates comfort of travel such that choosing the highest rated of these car parks would push the traffic system towards a minimized congestion state and consequently a more balanced car park usage.

It may be unlikely that every user follows the most recommended car park. However, because the disclosed method takes into consideration past user decisions, it is in fact not required that every user follows the recommendation from the Adaptive Load Balancer. In an embodiment, as long as there are enough users who can be controlled, the users can be managed by the present disclosure in such a way that the traffic imbalance on the road network can be balanced. Studies have shown that as long as 30% of the user population follows recommendations, a significant improvement in traffic distribution can be obtained.

The output of the third model 114 and the decision of the current user becomes part of the past decision data for input into the disclosed method on a future query. Therefore, the disclosed method may be capable of learning and adjusting the models disclosed herein to provide more accurate recommendations of available car parks within the geographical area. For example, the second model 112 may be adjusted to determine which model, e.g. which regression model, to use.

In an embodiment, there is provided a computer program product residing on a non-transitory computer readable storage medium of a system as disclosed herein, the storage medium having machine-readable instructions stored thereon which, when executed by the system, cause the system to perform the disclosed method. The computer program product may reside on the storage medium of a server that is remote from the user device. When a new user wishes to use the disclosed method, the user may use a client application or application programming interface located on the user device to communicate with the disclosed system or web application. The client application may be configured to receive user queries, communicate the user queries to the system, receive car park recommendations from the system, and present the recommendations as score card 140. Therefore the prediction software of the system as disclosed herein may be run on a remote server or hosted on a cloud, and the user device may retrieve the prediction result via the client application.

Claims

Patent claims

1. A method of generating an improved recommendation of at least one available car park amongst multiple car parks within a geographical area in response to a query of a current user received by a user device, wherein the method is executed in a system remote from the user device, the method comprising:

retrieving a query data feed comprising the query of the current user and a query of at least one other current user, each current query comprising: an intended parking location in a specified time window in the future defined by each current user, wherein the intended parking location is within the geographical area;

retrieving a past decision data feed comprising a final parked location of at least one past user;

analysing car park data of at least one car park within the geographical area, using a first model implemented by the processor, in order to predict car park occupancy for the specified time window in the future, wherein the car park data comprises current and historical car park occupancy retrieved from at least one public database;

analysing the query data feed and the car park occupancy predicted by the first model, using a second model implemented by the processor, in order to refine the prediction of the car park occupancy from the first model; analysing the past decision data feed and the refined prediction of the car park occupancy from the second model, using a third model implemented by the processor, in order to generate a recommendation of at least one available car park within the geographical area that causes the current user and the at least one other current user to be substantially uniformly distributed within the geographical area.

2. The method of claim 1, wherein the first model comprises an artificial neural network.

3. The method of any preceding claim, wherein the second model comprises a regression model.

4. The method of any preceding claim, wherein the third model comprises a statistical model or a traffic simulation.

5. The method of any preceding claim, wherein the geographical area is defined by a predetermined distance from the current user's intended parking location.

6. The method of any preceding claim, wherein the car park data is retrieved from the public database at a predetermined frequency .

7. A computer program product residing on a non-transitory computer readable storage medium of a system, the storage medium having machine-readable instructions stored thereon which, when executed by the system, cause the system to perform the method of any preceding claim.

8. A system of generating an improved recommendation of at least one available car park amongst multiple car parks within a geographical area in response to a query of a current user received by a user device remote from the system, the system comprising:

a database server configured to receive and store: car park data of at least one car park within the geographical area, the car park data comprising current and historical car park occupancy retrieved from at least one public database; query data comprising the query of the current user and a query of at least one other current user, each current query comprising: an intended parking location in a specified time window in the future defined by each current user, wherein the intended parking location is within the geographical area; past decision data comprising a final parked location of at least one past user;

an analytical server configured to analyse the car park data of each car park using a first model, in order to predict car park occupancy for the specified time window in the future,

the analytical server further configured to analyse the query data and the car park occupancy predicted by the first model, using a second model, in order to refine the prediction of the car park occupancy from the first model, the analytical server further configured to analyse the past decision data and the refined prediction of the car park occupancy from the second model, using a third model, in order to generate a recommendation of at least one available car park within the geographical area that causes the current user and the at least one other current user to be substantially uniformly distributed within the geographical area.

9. The system of claim 8, further comprising a web application configured to provide to the user device a score card listing the multiple car parks in order of recommendation, wherein the recommendation is based on car park occupancy and the distribution of the current user and the at least one other current user within the geographical area.

10. The system of claim 9, wherein the list of multiple car parks are colour coded based on the order of recommendation.