Reducing energy consumption in delivering services to recipients at delivery addresses

Abstract

A recipient of a service, such as a recipient of a package, usually carries a mobile device. A recipient has installed an application on a mobile device (1110) that is linked to a delivery agent (1105) and a time prediction system (1115). The identification code is assigned to the mobile device (1110). The application periodically sends its location along with the identification code to a time prediction system (1115). When a delivery agent (1105) receives an order to deliver a parcel to a recipient, the delivery agent (1105) sends a request (1130) to a time prediction system (1115) requesting the recipient for a predicted presence core period of the delivery address. The time prediction system (1115) sends a predicted presence core period of the recipient at the delivery address to the delivery agent (1105). The delivery agent (1105) is able to schedule a delivery (1140) to the recipient with the help of the predicted core period.

Description

Technical Field

[0001] The present invention relates to determining a core presence time of at least one service recipient at a delivery address. Furthermore, the present invention also relates to a data processing system comprising means for performing the above method, and to a computer program product and a computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to perform the above method. Background Art

[0002] A variety of methods are known for predicting whether an individual is located at a particular location within a specific time period. The results can be effectively notified to the terminals of home delivery companies to optimize product delivery times and avoid unnecessary trips, thereby reducing economic and environmental damage.

[0003] Description of Related Art

[0004] For example, predictions can be made based on data representing the energy consumption of a predetermined location (such as a home, workplace, etc.). However, this method lacks functionality and reliability. In fact, the results of specialized studies show that a quarter of the test households were excluded from the study because they could not establish behavioral characteristics due to insufficient energy consumption fluctuations. Energy use does not necessarily reflect user activity or presence status, and automatic devices, running washing machines, or heaters controlled by thermostats can distort the results. Participants need to fill out a detailed private life questionnaire and require the assistance of energy experts to draw conclusions. In addition, it is not easy to obtain consumer energy use data, and it cannot be obtained in real time unless one has a smart home. Consumers are not yet accustomed to sharing energy data. Despite the innovative research direction, these problems have hindered practical applications. This research also does not employ any data protection or security technologies.

[0005] Another method uses the data shared by a user when accessing a specific location to determine peak hours and stay durations. In particular, it is possible to determine the level of activity of a location at the current moment relative to the normal activity level. Furthermore, based on an estimation of the user's access pattern over the past few weeks, it is possible to determine the duration that a user usually stays at a certain location. According to this method, results are generated for public places or stores. Precise user location data is collected and stored, and can be obtained if hacked. In other words, sensitive information such as the streets a user passes through, the stores visited, the place of residence, and habit preferences may be illegally exploited.

[0006] According to other methods, the user's location is predicted based on mobile positioning data. First, the user's points of interest are identified through density clustering, and then the semantic attributes of the points of interest (home or workplace) are determined based on time assumptions. Finally, the future location is predicted through a decision tree model (trained based on the user's historical positioning data). This method only generates a weekly schedule and cannot provide real-time updates or optimizations during the delivery process. In addition, this method does not measure the distance between the user and the delivery address, nor does it remove time windows from the plan where it is impossible to arrive on time. Except for the user's authorization to use positioning in the application, this method does not use privacy and security technologies such as encryption or verification mechanisms, and the data processing is associated with the actual map data (latitude and longitude coordinates of the location).

[0007] Purpose of the Invention

[0008] Therefore, the present invention aims to solve or at least alleviate some of the above problems and reduce energy consumption when delivering services to recipients. Summary of the Invention

[0009] This object is achieved by the invention as described in the independent claim. Preferred embodiments are described in the dependent claims.

[0010] The present invention achieves the object of the present invention by a method. The steps of the method are described in detail below. The steps do not need to be performed in the exact order described. In addition, the method may include other steps not explicitly stated.

[0011] We propose a method for reducing energy consumption in delivering a service to a recipient of the service at a delivery address or geographic location.

[0012] Typically the recipient is a natural person, but it is also conceivable that it is a robot or machine that needs to have parts replenished or replaced.

[0013] A delivery address is typically a person's home or workplace, and includes general information such as the state / province, city, postal code, and street number. For high-rise buildings, this information may include a floor number, suite number, or room number. However, in remote areas without postal addresses, geographic coordinates in some reference system may be used to indicate the delivery location.

[0014] In most real-world scenarios, a service refers to the delivery of a package or goods. However, it can also include services provided by craftspeople, mechanics, nurses, and so on. Broadly speaking, a service is an activity that requires the simultaneous presence of at least two people at the delivery address or location. More broadly, a service can be delivered by a robot or drone, requiring only the simultaneous presence of the robot / drone and the recipient, a natural person or machine, at the delivery address or location.

[0015] The method requires multiple steps. The mobile device of the receiving party needs to send data at least including the identification code corresponding to the receiving party and the location representation data of the mobile device to the time prediction system at fixed time intervals.

[0016] Typically, the mobile device needs to enable a positioning sensor. The sensor is usually a built-in GPS module, but it can also be any other positioning means. For example, through mobile network base station triangulation, or using Wi-Fi / Bluetooth for position determination.

[0017] Generally, an application with data sending permission needs to be running on the mobile device of the recipient. After installation, the program sends data to the time prediction system at fixed intervals, such as every 15 minutes or every minute, etc. The sending frequency can be increased when the device is moving and decreased when it is stationary.

[0018] The time prediction system receives and stores the position characterization data, and stores it together with the identification code and timestamp of the recipient.

[0019] An entity receives an order to deliver a service to a recipient at the delivery address or geographical location. The entity can be a path planning system of a logistics company, a courier, or a scheduling system of a nursing service agency or an online store.

[0020] The order contains at least the delivery address or geographical location and data identifying the recipient. The data of the recipient can be, for example, the recipient's name or a customer number in an online store.

[0021] The entity further receives the identification code of the recipient or calculates the identification code of the recipient based on relevant information about the recipient. This code can be generated by an application on the recipient's mobile device, or generated by the logistics system and transmitted to the application, or generated by an online store where the recipient purchases goods. If a specific algorithm is used, each participating party can independently calculate the identification code based on the relevant information about the recipient that they have.

[0022] After receiving the service order, the entity sends a request to the time prediction system, asking the system to provide the predicted presence core period of the recipient at the delivery address or geographical location to the entity. The request contains at least the delivery address / geographical location and the identification code corresponding to the recipient. It can also contain the earliest feasible time and date when the entity can provide services to the recipient at the delivery address or geographical location.

[0023] After the time prediction system responds to the request, it provides the entity with the predicted core presence period of the requested recipient, enabling the entity to deliver the service to the recipient at the delivery address or geographical location during the predicted core presence period of the recipient at the delivery address. Usually, the system will return the predicted core presence period for the next few days, such as the predicted core presence period for the next 2 - 7 days, preferably the predicted core presence period for the next 2 - 5 days, and most preferably the predicted core presence period for the next 2 - 3 days. For example, in the door-to-door delivery scenario, these predictions can be limited to preferred receiving time periods such as 08:00 - 21:00 from Monday to Saturday, etc.

[0024] If the request includes the earliest feasible time and date for the entity to provide service to the recipient at the delivery address or geographical location, the predicted values provided will be calculated starting from this earliest time and date; otherwise, the provided predicted core period starts from, for example, the moment the request is received.

[0025] This method can significantly reduce ineffective delivery trips, saving up to 45% of the time and energy consumption in the delivery trips on average.

[0026] By significantly increasing the probability of the delivery agent meeting the recipient at the delivery address or geographical location, this method effectively avoids the situation where customers wait in vain for the service. Practice has proved that such situations are the main factors causing strong dissatisfaction among customers.

[0027] Regarding only the time prediction system, the proposed method will operate as follows from the perspective of this system. The method for calculating the predicted core presence period of the service recipient at the delivery address or geographical location needs to perform the following steps from the time prediction system's view.

[0028] The time prediction system receives at regular time intervals from the mobile device of the recipient at least the identification code corresponding to the recipient and the location characterization data of the mobile device.

[0029] The time prediction system stores the location characterization data together with the identification code and timestamp of the recipient.

[0030] When the entity needs to perform service delivery, the time prediction system will receive a request sent by the entity asking for the predicted core presence period of the recipient at the delivery address or geographical location.

[0031] The request includes at least the delivery address or geographical location and the identification code of the recipient, and may also include the earliest feasible time and date for the entity to provide service to the recipient at the delivery address or geographical location.

[0032] After receiving the request, the time prediction system provides the requested predicted presence core period to the entity, enabling it to deliver the service to the recipient during the predicted presence core period of the recipient at the delivery address.

[0033] For this purpose, the present invention requires that the time prediction system includes means for performing the above method.

[0034] Furthermore, the object of the present invention can be achieved by a computer program which includes instructions that, when executed by a computer, enable the computer to implement the method.

[0035] For this purpose, a readable medium storing the computer program according to the foregoing claims can be provided on the market.

[0036] Correspondingly, the present invention requires a computer program for a service recipient's mobile device, which sends at least the identification code of the recipient and the location characterization data of the mobile device to the time prediction system that executes the method at fixed time intervals.

[0037] From the perspective of an entity involved in delivering a service to a recipient at a delivery address or geographical location, the method includes the following steps: the entity receives an order to deliver a service to the recipient at the delivery address, and the order includes at least the delivery address or geographical location and the name of the recipient.

[0038] The entity will also receive the identification code of the recipient, or calculate the identification code of the recipient based on the relevant information of the recipient.

[0039] After receiving the service order, the entity sends a request to the above-mentioned time prediction system to obtain the predicted presence core period of the recipient at the delivery address or geographical location. The request includes at least the delivery address or geographical location itself and the corresponding identification code of the recipient, and may also include the earliest feasible time and date for the entity to provide the service to the recipient at the delivery address or geographical location.

[0040] After sending the request, the entity receives from the time prediction system the requested predicted presence core period of the recipient at the delivery address or geographical location, thereby enabling it to deliver the service to the recipient during the predicted period of the recipient at the delivery address.

[0041] Preferably, the location data of the mobile device of the recipient is subject to conversion processing before being sent to the time prediction system, so that the time prediction system cannot determine the location of the mobile device on the earth based on this. However, the converted location data still allows the time prediction system to measure the relative distances between different locations within a preset distance range of the mobile device. Correspondingly, the service delivery address or geographical location for the delivery service also needs to be subjected to the same conversion, which can be done by the entity before sending it to the time prediction system in the request, or completed within the time prediction system.

[0042] This will ensure that the time prediction system cannot learn the exact location of the recipient at any point in time. The system can only predict when the service recipient will appear near the service delivery address or geographical location, but only relative location information is known, rather than the absolute location on the earth's surface. The system can infer whether the recipient is at home, but cannot determine whether this home is in the United States or Europe.

[0043] This step further enhances compliance with data protection regulations.

[0044] Preferably, the conversion of the location characterization data of the mobile device is achieved by means selected from the following:

[0045] -- Performing geohashing on the location data and deleting an appropriate number of leading digits in the geohash code, preferably deleting the first three digits;

[0046] -- Deleting an appropriate number of leading digits in the geographical coordinates of the location data, preferably deleting the integer geographical degree value while retaining the geographical minutes and seconds value.

[0047] Therefore, whether it is the recipient's mobile device sending the delivery address, geographical location, and more location information to the time prediction system, or any entity participating in the service delivery, only the delivery address and location data in the converted format are sent to the time prediction system. This prevents the time prediction system from reconstructing the actual location of the recipient on the earth.

[0048] To ensure that the time prediction system cannot reconstruct the recipient's identity, the identification code corresponding to the recipient needs to meet the following conditions: the system cannot reconstruct the recipient's name and delivery address. For this purpose, an identity code can be assigned to each service recipient. Preferably, the identification code is generated by hashing the data identifying the recipient, and more preferably by hashing the recipient's email address. If the entity sending the request to the time prediction system knows the recipient's email address and the identification code generation algorithm, the entity can calculate the identification code by itself without obtaining it from the recipient's mobile device. The identification code can also be generated by any entity participating in the service delivery or the recipient's mobile device. Regardless of the generating party, this code needs to be distributed to all service delivery participants except the time prediction system.

[0049] In order to increase the probability of meeting the recipient at the delivery address or geographic location, and thus improve the success rate of service delivery, more members of the recipient's family can be included in the service scope. Specifically, the group of mobile devices corresponding to the recipient group needs to regularly send the following data to the time prediction system: the corresponding identification of the individual recipient or the identification code shared by all recipients in the recipient group, and the location representation data of the individual recipient's mobile device. This requires that new potential service recipients also need to regularly appear at the delivery address or geographic location of the main recipient. For example, other family members or cohabitants of the recipient can participate in this process. Based on this, the time prediction system will calculate the predicted core presence time of any recipient in the group at the delivery address or geographic location.

[0050] When this preferred embodiment of the present invention is adopted, energy consumption during the service delivery process can be further reduced.

[0051] Preferably, the time prediction system calculates the probability that a mobile device will be within a predetermined distance of the delivery address or geographic location at a specific moment. This predetermined distance should ensure a high probability of contacting the recipient at the time of delivery. The predetermined distance can be set, for example, to 10m, 20m, 30m, 40m, or 50m.

[0052] The simplest way to calculate this probability is to count the relative frequency with which the recipient appears at the delivery address or geographic location.

[0053] The predicted core period of the recipient's presence at the delivery address or geographic location is preferably determined by comparing the calculated probability with a preset threshold probability, which may be set to, for example, 80%, 90%, or 95%.

[0054] More accurate, flexible, and reliable calculation methods include using machine learning or deep learning methods to predict core attendance periods, preferably using recurrent neural network models, long short-term memory (LSTM) models, such as the PewLSTM with weather-aware gating mechanism, or Transformer-based models. These improvements can help further reduce energy consumption in the service delivery process.

[0055] To further improve the probability of locating the recipient at the postal address or geographic location during delivery, the time prediction system can be provided with additional information. This additional information is most preferably weather data. The time prediction system will then calculate the core hours during which the recipient will be present at the delivery address or geographic location based on or taking into account the additional information.

[0056] To enhance the data security of the described method, several measures can be implemented. For example, a verification code can be used to confirm that the mobile device is authorized to transmit location-representing data to the time prediction system. This verification code must be exchanged in advance between the time prediction system, the entity, and the mobile device.

[0057] Another solution to enhance data security is to send the location representation data of the mobile device to the time prediction system through a proxy server. In this way, the time prediction system cannot even know the IP address of the receiving mobile device.

[0058] The following solution can also be adopted: The time prediction system generates an event token after receiving a request from an entity such as an online store. This token will be bound to a specific delivery event. After obtaining the event token, the entity or its entrusted delivery agent, such as a package delivery person, can use this token to initiate subsequent requests to the time prediction system for this delivery event. The event token can also be transferred from the online store to the delivery agent. When the service is successfully delivered to the recipient at the delivery address, the event token becomes invalid.

[0059] This method determines the predicted core presence period of the recipient at the service target delivery address or geographical location. If the actual presence status of the recipient at the delivery address is verified immediately before the service delivery, invalid delivery attempts can be further avoided. The above verification can be achieved by the entity sending a corresponding request to the time prediction system.

[0060] In an alternative solution, the time prediction system continuously monitors the distance between the location of the mobile device and the delivery address / geographical location, and then determines whether the mobile device is at the target location or can reach the target location before the predicted core presence period. When it is difficult for the mobile device to reach the target location within the predicted period, the system will update the predicted core presence period. The updated data can be actively pushed to the entity (push service), or the entity can regularly request an update from the system (pull service).

[0061] In other words, the time prediction system ensures that the system does not predict a core presence period during which the mobile device cannot reach the fixed converted location within the predicted period by comparing the distance between the dynamically converted location of the mobile device to be measured and the fixed converted delivery address / geographical location.

[0062] To achieve the object of the invention, we also propose a method comprising the following steps:

[0063] · Collect data at a collection point;

[0064] · Anonymize the data at the collection point;

[0065] · Transmit the anonymized data to a data room;

[0066] · Augment the anonymized data in the data room, for example, obtain personalized recommendations or insights from it through a deep learning model;

[0067] · Retrieve the personalized recommendations or insights using a cryptographic key, where the possibility of retrieving the personalized recommendations or insights is restricted in time.

[0068] This method can generate insights while meeting data protection requirements, and these insights are valuable for a variety of application scenarios, including reducing energy consumption in the service delivery process. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Other objects and advantages of the present invention will become apparent from reading the specification and claims and in conjunction with the accompanying drawings. To understand the present invention more comprehensively, please refer to the following embodiments described in conjunction with the drawings. The problem-solving solutions are not limited to the illustrated embodiments, and the same reference numerals in the drawings represent the same or functionally equivalent elements. The specific illustrations are as follows:

[0070] Figure 1 A flowchart of a method for calculating a user's predicted presence core period according to an embodiment;

[0071] Figure 2 A schematic diagram of the method steps executed by a time prediction system according to an embodiment;

[0072] Figure 3 A schematic diagram of an overview of information exchange among a user, an application operator, and a time prediction system according to an embodiment;

[0073] Figure 4 A flowchart of a method for requesting a user presence prediction according to an embodiment;

[0074] Figure 5 A schematic diagram of an overview of information exchange among a user, an application operator, and a time prediction system when a user generates an event according to an embodiment;

[0075] Figure 6 A flowchart of a "Delivery Location API" and an "Event Token API" according to an embodiment;

[0076] Figure 7 Another flowchart of a "Delivery Location API" and an "Event Token API" according to an embodiment;

[0077] Figure 8 A flowchart of a "Core Period API" and a "Backup Check API" according to an embodiment;

[0078] Figures 9A to 9C Schematic diagrams of an RNN unit, an LSTM unit, and a PewLSTM unit respectively;

[0079] Figure 10 A schematic diagram of a PewLSTM architecture;

[0080] Figure 11 A schematic overview diagram of an embodiment of the method;

[0081] Figure 12 FIG. 1 is a schematic diagram showing an overview of another embodiment of the method. DETAILED DESCRIPTION

[0082] This solution enables the user (i.e., consumer) to be informed of the core hours of presence at the delivery address in compliance with the General Data Protection Regulation (GDPR). To this end, operators, such as application operators 22 (see Figure 3 ), will be combined with the time prediction system 3 (see Figure 2 and Figure 3 ) to interact as described below. The application operator 22 may be, for example, an online store.

[0083] The time prediction system 3 is configured to provide a presence core period prediction (DCP) that complies with data protection requirements. Therefore, according to the present disclosure, the time prediction system 3 can also be regarded as a DCP system.

[0084] like Figure 1 and Figure 2 As shown, in one embodiment, a method 100 for calculating a user's predicted core presence period 7 includes: at step S101, forwarding the user's delivery address information data 1 and identification data 2 to a time prediction system 3. Specifically, these information data are sent to the time prediction system 3 by an application operator via a forwarding element, such as the application operator 22. The delivery address data 1 and identification data 2 may be referred to as first input data 4.

[0085] In step S102, the application user's location information data 5, along with their identification data 2, is forwarded to the time prediction system 3. The application user is the recipient of the service. The location data 5 and identification data 2 can be referred to as second input data 6. These data are forwarded sequentially, meaning they are sent multiple times at different times or at regular intervals. In step S103, the time prediction system 3 checks the authentication status of the application operator.

[0086] Then, in step S104, the first input data 4 and the second input data 6 are processed to calculate the predicted core presence period 7 of the application user. The prediction is based on at least the delivery address information data 1 and the location information data 5 of the application user, and the predicted core presence period 7 is generated as output data 8. These predicted values represent a highly likely time period DCP, i.e., a time range in which at least one person will be present at the delivery address.

[0087] The input data processing includes a conversion step (S105). Specifically, the delivery address information data 1 of the application user is converted into the coordinate position of the address on a relative map. The relative map refers to any coordinate system that cannot identify the actual location on the earth through its coordinates, but whose coordinates allow measuring the distance between different coordinate points on the relative map within a preset distance range.

[0088] After converting the geographical coordinates of application user 22 into relative map coordinates, these coordinates are converted into address geohash values 9 through a hash function. The location information data 5 of the user is also first converted into relative map coordinates and then converted through a geohash function 10. It should be noted that any attempt to reverse the geohash value can only obtain fictional information that has nothing to do with the user's true personal information, delivery address, and location data, thereby enhancing the data protection and security of application users. As will be described in detail later, the DCP system employs multi-layer data protection and security technologies to meet global data protection standards.

[0089] It should be noted that the DCP system adopts a normative analysis method, which means providing automated action suggestions about the future based on the facts and probability-weighted predictions of predictive analysis. Predictive analysis includes various statistical techniques such as data mining, predictive modeling, machine learning, and artificial intelligence, which analyze current and historical facts to predict future or unknown events.

[0090] As input data (the first and second input data), the mobile location data of the application user can be regarded as core data. However, other data can also be additionally used, such as regional information data including observed weather conditions or forecasts.

[0091] In another embodiment of the present invention, method 100 includes providing additional information 15 as the third input data 16 to the time prediction system 3, where the additional information data 15 particularly includes meteorological data. By automatically analyzing this information, the impact of such conditions (such as weather) on the address presence pattern can be understood and the prediction can be improved. It should be noted that the regional information data is not limited to meteorological information, and other data sources can also be added, including but not limited to traffic conditions or local holiday information.

[0092] Figure 2 Schematically shows the processing process from input data 4, 6, 16 to output data 8. The first input data 4 in one embodiment at least includes the delivery address information data 1 and identification data 2 of the application user. These data only need to be provided to the time prediction system 3 or the DCP system once. The second input data 6 at least includes the location information data 5 and identification data 2 of the application user. These data are sequentially provided to the time prediction system 3. In addition, the third input data 16 is also provided to the time prediction system 3, and these data particularly include additional information data 15 such as meteorological information.

[0093] The time prediction system 3 includes a conversion module 12 for converting the delivery address information data 1 and location information data 5 received by the application operator into hash values. The conversion module 12 exists in the form of a software development kit (SDK). Specifically, the delivery address information data 1 of the application user is converted into the delivery address position on the relative map, forming a geographical hash value related to the delivery address (i.e., address geographical hash 9), and the location information data 5 of the application user is converted into the coordinates of this location on the relative map, forming a geographical hash value related to the location (i.e., location geographical hash 10). This process is achieved through the hash function module 11. The coordinates are then used to generate a geographical hash value through hash function operations. In other words, when converting the location information data 5 and delivery address information data 1 of the application user into the delivery address and location coordinates on the relative map, the precise location information of the application user on the earth is removed from the address geographical hash 9 and location geographical hash 10. The application user number undergoes a hash operation to generate an identification code.

[0094] In some embodiments, the system periodically compares the coordinates of the address geographical hash 9 and the location geographical hash 10 on the relative map to obtain comparison result data 14 for predicting the core presence time 7 of the application user. For this purpose, the time prediction system 3 also includes a comparison processing module 13. Specifically, the system calculates the distance between the delivery address position on the relative map and other access positions to obtain the comparison result 14.

[0095] This periodic comparison provides a basis for predicting the core presence time within a specific time range. Through continuous comparison, continuous learning of the core presence time can also be achieved, and the update of the core presence time is ensured when a change in the behavior pattern is detected.

[0096] To further enhance data security, the core presence time is calculated only when an entity issues a prediction request, and this calculation is repeated only until the service delivery, i.e., the home delivery, is completed. In other words, only when the service delivery is required, the presence situation of the application user at the specified delivery address is predicted. If no delivery event is scheduled, the core presence time is not calculated.

[0097] Figure 3 Shows an application embodiment of the method 100. In Figure 3 the illustrated embodiment, the first input data 4 and the second input data 6 are forwarded by the application operator 22 to the time prediction system 3.

[0098] The DCP system can be integrated through an Application Programming Interface (API). To fully leverage its capabilities, the DCP needs to be integrated into an agile system capable of handling real-time data. As an additional feature, the DCP can also be seamlessly embedded into the integrated system in the form of a white-label solution. In the last-mile logistics application scenario, the core systems that can be connected to the DCP include the order processing system of online retailers, as well as the transportation management system, route planning system, and path optimization system of delivery services.

[0099] Reference Figure 3 , after the application user (or consumer) authorizes the use of location information in the APP of the application operator 22, the application operator 22 will automatically send the delivery address information data 1 and the user ID number 2 of the service recipient (application user) to the time prediction system 3 (i.e., the DCP system) together. In addition, the location information data 5 and ID number 2 of the application user will also be sent to the DCP system 3 by the application operator 22 in sequence. Note that in the figure, the sequential transmission of data is represented by a dashed line, while a single transmission is identified by a solid line.

[0100] When any message is directed to the DCP API system, the first step is authentication. New application operators will receive a verification code during the DCP integration process, and this verification code needs to be sent with each message to the DCP API. Specifically, method 100 includes the time prediction system 3 verifying this verification code when receiving at least the first and second input data 4, 6. Messages without a valid verification code will not be processed, and incoming data without a verification code will be automatically rejected.

[0101] Figure 4 The flowchart of an embodiment of method 200 for requesting the presence prediction of an application user is shown. This method is particularly applicable to requesting the presence prediction of at least one user at a specific delivery address. This function is especially useful when there are multiple users at the same delivery address.

[0102] In step S201, the predicted core presence period 7 of the application user is calculated. This predicted value is generated according to the aforementioned method 100, so the detailed steps will not be elaborated here. In step S202, the application user or an entity participating in service delivery generates an event 23. The event 23 can be, for example, a commercial order. Subsequently, in step S203, usually, the entity related to service delivery forwards the presence prediction request 17 (see Figure 5 ) to the time prediction system 3.

[0103] In some embodiments, the presence prediction request 17 includes the delivery address 18, the application user number or identifier, the application operator verification code, and event information data, where the event information data is particularly used to declare that the event 23 is occurring.

[0104] In step S204, the time prediction system 3 creates an event token 19 and sends an information request 20 regarding the core presence period 7 of the application user together with this event token 19 to the time prediction system 3 (S205).

[0105] Finally, in step S206, the system analyzes the validity of the event token 19 and sends result information 21, which contains the previously calculated core presence period 7 of the application user at the requested delivery address. Note that the result information 21 neither contains any personal data of the application user nor involves map data associated with the actual geographical location on Earth.

[0106] In some embodiments, the method includes authenticating and hashing the presence prediction request 17 by the time prediction system 3, and this operation is performed in step S207.

[0107] When the event 23 generated by the application user is completed, the method may include a step of deleting the event token 19 (S208).

[0108] Figure 5 An application embodiment of the method 200 is shown.

[0109] When a specific event 23 (such as an order with home delivery desired) occurs, the event occurrence information, together with the corresponding customer number R identification and the delivery address, is forwarded to the DCP system 3 through the presence prediction request 17.

[0110] After the request 17 is successfully authenticated and hashed, the DCP system 3 automatically creates and returns an event token 19. With this token, an information request 20 can be sent to the DCP API system 3 to obtain the result information 21. If the event token 19 is valid, the DCP system 3 will reply with the result information 21. The result only contains the predicted core presence period 7 of the delivery address and does not involve the delivery address itself and the application user number 2, that is, the result 21 does not contain any personal data. In other words, in the complete interaction between the application operator 22 and the DCP system 3, the system never knows the name of the application user / consumer. Only the authorized operator 22 who initially provided the application information can associate the result with the delivery address because it knows the specific user corresponding to the event token 19.

[0111] When the purpose of the event token 19 is achieved (such as the goods have been delivered), the information will be automatically transmitted to the DCP system 3. The system will immediately delete the event token 19 and return a deletion confirmation. The deleted token cannot obtain any information, and all requests carrying the deleted token will be automatically rejected.

[0112] It should be noted that the software development of the DCP system 3 exceeds all global data protection standards to ensure global applicability and recognition. Therefore, the principle of minimizing personal data is adopted, that is, only the most basic personal data necessary for the application users of the application operator 22 of the DCP system 3 to achieve their consent purposes is used when using the mobile application of the DCP system 3.

[0113] As mentioned above, this solution significantly improves data protection and security. The standards adopted by the DCP system 3 when processing mobile location data are much stricter than the current location-based services (LBS). Through a number of measures described below, the DCP system 3 sets a new benchmark for data protection and security.

[0114] 1. Verification code

[0115] With the verification code, only the authorized application operator 22 can interact with the system regarding the events 23 of its own application users.

[0116] 2. Anonymization processing & hashing processing of all input data

[0117] The system cannot process or identify name information, so it never knows the user's name. The application user number, delivery address, and location data are all hashed, and there is no plaintext data in the system.

[0118] 3. Relative map (not real address & not real location)

[0119] The delivery address and location data 1, 5 exist in the system in the form of geographical hash values 9, 10. These geographical hash values 9, 10 are stored as coordinate points on the relative map. Since it does not contain real map data, even if the system is invaded and the hacker successfully performs "reverse hashing" processing, these data can only show the coordinates on the relative map (without specific address, city, or country information) and cannot be actually utilized. Different from the mainstream LBS, this solution does not track the user's travel route, visited addresses, or shops, and the DCP system 3 does not analyze purchase preferences and is never used for any form of user / consumer behavior intervention.

[0120] 4. No absence confirmation or prediction

[0121] The DCP system 3 does not predict the absence period, and the result does not constitute an absence confirmation. Usually, the address period can be divided into: frequent presence period, frequent absence period, and variable behavior period. This system does not distinguish or analyze the frequent absence period and the variable behavior period, and only predicts the core presence period when there is usually always someone at the delivery address and dynamically updates these periods.

[0122] 5. Location usage authorization

[0123] The DCP system 3 only processes application user data used at confirmed continuously authorized locations by users who use applications of application operators 22 integrated with DCP and are aware of the purpose of use during the confirmation process.

[0124] 6. Return Results of Event Token / No Address & Personal Data / Time Limit Control

[0125] The results generated by DCP can be obtained with a valid event token 19. This token will be sent to the party that initially provided the application user information. The processed results (obtainable only with a valid event token 19 before the purpose of event 23 is achieved, e.g., the package has been delivered) do not contain the address or personal information of the application user. Only the authorized application operator 22 that applied for the event token 19 can associate the results with the delivery address during the duration of event 23 (e.g., during delivery).

[0126] Figure 6 、 7 Figure 8 shows a schematic diagram of the operation process of the above method in some embodiments. These flowcharts represent application programming interfaces (APIs), i.e., public endpoints for verifying identities, collecting and forwarding data, and returning information.

[0127] Figure 6 In the illustrated embodiment, the delivery location API first collects the verification code and verifies its validity. If it is invalid, the request is rejected; if it is valid, the authentication is completed, and then the customer ID 2 is hashed and stored in persistent storage. The coordinates relative to the map are converted and synchronously stored in the form of a geographical hash value.

[0128] Figure 7 The illustrated event API verifies the validity of the verification code and the presence of the user. Invalid requests are rejected, and valid requests pass the authentication. Subsequently, the longitude and latitude of the delivery address relative to the map are converted into a geographical hash value 9, which is stored in persistent storage together with the hashed customer ID and specific event 23 information. After calculating the distance based on the stored data in the relative map, the delivery address distance data is input into the model for training.

[0129] Figure 8 The illustrated core period API verifies the validity of the event token. Invalid requests are rejected, and valid requests obtain the predicted core presence period from the trained model and return it in the form of an array. If the event is not completed, the system will repeatedly obtain and predict the core presence period 7 to review the prediction results and respond to changes in user behavior.

[0130] In some embodiments, when the matching relationship between the location of the delivery address geo-hash 9 and the location of the relative map location geo-hash 10 changes within a predetermined time period, the core presence time 7 of the application user can be updated. Capturing changes in presence behavior and updating presence predictions accordingly is crucial for adapting to situations such as changes in working hours or habits. For example, changes such as someone starting to work the night shift or someone starting to take a walk in the park at noon will affect presence predictions. These changes are particularly important for specific services related to consumer presence, such as home delivery.

[0131] As will be described in detail below, to predict the core presence time 7 of an application user, machine learning or deep learning methods are needed to process the data. By training a machine learning or deep learning model, the presence probability of the user in a specific time period is calculated. Specifically, the comparison result data 14 obtained by calculating the distance between the delivery address location and other access locations on the relative map is input into a machine learning or deep learning algorithm to perform core presence time prediction.

[0132] The model proposed in this solution aims to predict the core presence time window of an application user at the delivery address, partially based on historical data. Such problems are called "time series prediction". There are various existing models to solve this problem. For example, the univariate "autoregressive moving average (ARMA)" model for single time series data (combining the autoregressive (AR) and moving average (MA) models), or the univariate "autoregressive integrated moving average (ARIMA)" model considering differencing operations, are both traditional time series prediction methods. With the popularization of machine learning (ML) and deep learning (DL) technologies, newly emerging are deep learning-based methods. The most successful algorithms among them include the recurrent neural network (RNN) and its variant the long short-term memory network (LSTM). "PewLSTM" is a new periodic weather-aware LSTM model that predicts future parking behavior by combining historical data, weather, and day-of-week factors, with its accuracy improved by about 20% compared to the existing best parking behavior prediction methods. Since the core presence time of a predictor at a specific delivery address is affected by day-of-week and weather factors, the PewLSTM model is applicable to this solution.

[0133] Time series data is a dataset of continuous quantities collected at uniform time intervals, used to track the development trend of data. The process of predicting the future evolution of this variable through statistical analysis and modeling is called time series prediction. When using a single feature, it is called univariate prediction, and when using multiple features, it is called multivariate prediction (more complex). Predicting the value of the subsequent single time step or multiple time steps is called single-step prediction and multi-step prediction respectively. As mentioned before, the difference between univariate and multivariate prediction lies in the number of features used to train the model, so only the dimension of the input vector needs to be changed. However, there are far more multi-step prediction methods than single-step prediction, mainly including: direct multi-step prediction, recursive multi-step prediction, and multi-output prediction. In direct multi-step prediction, an independent model needs to be trained for each prediction time step, which is different from recursive multi-step prediction where the single-step model is recursively used for iterative prediction. In multi-output prediction, the model is trained once to complete the prediction of the overall output in vector form.

[0134] RNN is one of the deep learning methods for time series prediction. A recurrent neural network is a deep learning model for processing sequential data (such as time series data), and its output depends on the previous elements in the sequence. RNN processes variable-length sequence inputs through an internal state (memory), which is called the recurrent hidden state.

[0135] We define x=(x1,x2,x3,…,x t ) as the input sequence, and h t as the recurrent hidden state at time step t. h t is updated through the following formula:

[0136] h t = σ(W x x t + W h h t-1 + b t )

[0137] where the non-linear activation function is denoted as σ, for example, it can be the logistic Sigmoid function, the hyperbolic tangent function, or the rectified linear unit (ReLU). W x , W h represent weight matrices, and b t is the constant bias.

[0138] Figure 9A shows the RNN cell.

[0139] Recurrent neural networks (RNNs) have difficulty learning long-term dependencies due to the "vanishing gradient" problem. The gradient is the partial derivative of a function with respect to its input and measures the impact of changes in the input on the output. The "vanishing gradient" problem occurs when the weight matrix becomes too small, causing the model to stop learning. Since long sequences imply multiple layers, the gradient is highly likely to vanish. The Long Short-Term Memory (LSTM) model effectively solves this problem.

[0140] Its variables include:

[0141] f t = σ(W fx x t + W fh h t-1 + b f )

[0142] i t = σ(W ix x t + W ih h t-1 + b i )

[0143] g t = tanh(W gt x t + W gh h t-1 + b g )

[0144] o t = σ(W ox x t + W oh h t-1 + b o )

[0145] c t = g t ⊙ i t + c t-1 ⊙ f t

[0146] h t = tanh(c t ) ⊙ o t

[0147] where ⊙ represents the Hadamard product (element-wise product), and f t represents the forget gate, which determines the information to be cleared from the LSTM memory. The input gate (which determines whether new information is stored in the LSTM memory) is represented by it. g t is the cell input activation gate through which the new candidate vector is stored in the LSTM memory. o tIs the output gate, which determines the value of the next hidden state. Cell state c t Stores the information of the previous time interval in the LSTM unit, and the hidden state h t The information of the previous cell is passed to the subsequent cell. The weight matrix W reflects the impact of the input change on the output, and the bias term b determines the deviation between the function output and the expected result.

[0148] Figure 9B Demonstrating the LSTM unit.

[0149] While LSTM models address the vanishing gradient problem, they are unable to learn periodic patterns and weather data. These issues are addressed in a periodic LSTM with a weather-aware gating mechanism, called PewLSTM. Figure 9C Showing the PewLSTM unit structure.

[0150] The difference between PewLSTM and typical LSTM is the addition of a gating mechanism for historical periodic observation data and meteorological data. Behavior at the same time yesterday, last week, and last month may affect the current state. Considering that human activities (such as work and study) often follow a biweekly pattern, this pattern is also taken into account. Variable h day ,h week 、h biweekly and h month Describe the hidden states that represent these behaviors respectively, and then combine these hidden states with the hidden state h of LSTM through a special weight gate δ t-1 Integrate into the following formula. The function of the weight gate is to continuously update the weight of each parameter.

[0151] h o =(W d h day +W w h week +W bw h biweekly +W m h mon +W t-1 h t-1 )

[0152] The meteorological characteristic vector e can be introduced t To capture the impact of weather information on on-site behavior, this vector is integrated into the forget gate, input gate, and output gate. The weather data at time step t is used as the input of the standard feed-forward layer and processed using the sigmoid activation function.

[0153] e t =σ(W e weather t t +b f )

[0154] The weighted hidden state h0 and the meteorological input e t are then integrated into f t , i t and o t (corresponding to the forget gate, input gate, and output gate respectively). However, the formulas for the hidden state h t and the cell state c t remain the same as those of the LSTM because the periodicity and meteorological information have been processed and stored through the modified variables f t , i t and o t as shown in the following equations.

[0155] i t = σ(W ix x t + W ih h o + W fe e t + b i )

[0156] g t = tanh(W gx x t + W gh h o + b g )

[0157] f t = σ(W fx x t + W fo h o + W fe e t + b f )

[0158] o t = σ(W ox x t + W oh h o + b o )

[0159] In some embodiments, for at least one user with a specific delivery address, different data processing techniques are used to perform serial repeated calculations on the predicted core presence period 7.

[0160] Combined with the calculation of the distance between the hash position data on the relative map and the hash delivery address, different technologies can be used for serial repeated calculations to achieve real-time optimization. In particular, when the system recognizes that the user cannot arrive at the delivery address on time, the corresponding predicted core presence period 7 will be deleted. Therefore, even if there are deviations in the core period predictions for the next few days, as long as there are no abnormal situations, the real-time optimization mechanism will automatically correct the errors. Abnormal situations include but are not limited to: small probability events such as the application user turning off the phone, the phone battery running out, the user losing the phone, or a server failure.

[0161] The loss function is used to evaluate the performance of the model during model training. It represents the error between the model output and the target variable (the variable to be predicted). The goal is to minimize the loss value, which means the degree of closeness between the predicted data and the actual data. Common loss functions include: mean absolute error (MAE) and mean squared error (MSE) for continuous output data, and binary cross-entropy (BCE) for binary classification data. Since the output is binary (presence prediction core period or absence prediction core period within a time range), this paper introduces the binary cross-entropy (BCE) loss function and its variants to improve the prediction effect of the trained model.

[0162] The formula for the binary cross-entropy loss function is as follows:

[0163]

[0164] where p i and y i represent the i-th scalar value of the model output and the i-th scalar value of the target value respectively. N represents the output size.

[0165] The binary cross-entropy loss function requires the use of the sigmoid activation function before the target layer because it compresses the output to between 0 and 1. Its working principle is: when the actual class value is 1, the second part of the formula fails, and vice versa. Since the target value p i is between 0 and 1, the logarithm of p i or 1 - p i ranges from negative infinity to 0. Therefore, the closer the predicted value is to the actual class, the smaller the loss value.

[0166] The weighted binary cross-entropy loss is an extended form of the binary cross-entropy loss, which punishes the misjudgment of positive examples by increasing the positive example weight. This method specifically punishes the mispredicted core presence period at a specific location. The formula for the weighted binary cross-entropy loss is as follows:

[0167]

[0168] where p i , y i and N are defined as before, and wi Represents the positive example weight.

[0169] Calculate the input data for predicting the core presence period 7 according to this method, that is, the distance between the location and the delivery address (as described above). Calculate the distance d between the delivery address position and other visited positions on the relative map through the two-dimensional plane distance formula. The formula is:

[0170]

[0171] Among them, (x1, y1) represents the coordinates of the first point on the relative map, and (x2, y2) represents the coordinates of the second point.

[0172] The x and y values can be obtained by converting the minutes and seconds of geographical longitude and latitude into x, y plane coordinates. If the geographical hashing algorithm is used, the x and y values can be obtained by decoding the Geohash value of the converted position data. Generally speaking, the method adopted must be applicable to the reverse restoration of the converted position data.

[0173] The distance is in meters. Use a Z value of 3 for outlier filtering, and set the lower limit of outliers to this threshold. A Z value of 3 means that any distance exceeding three standard deviations of the data distribution will be regarded as an outlier.

[0174] Meteorological data will be used as model input together with distance data. Meteorological features include but are not limited to air temperature, relative humidity, wind speed, and precipitation height. The data needs to be normalized by a MinMax scaler before being input into the model.

[0175] Regarding model training, it should be noted that a custom periodic LSTM with a weather-aware gating mechanism is implemented using the above-defined variables. For each batch size, the system captures the hidden states and weather information of the last day, week, bi-week, and month to calculate the next hidden state, which is then used to update the model parameters through backpropagation. Backpropagation refers to calculating the gradient of the loss function with respect to the network weights. When training a deep learning model, an optimizer, learning rate, number of training epochs, and batch size need to be configured: The optimizer is a function that searches for optimal parameters and weights to optimize the results; the learning rate is the step size used by the optimizer to minimize the loss function in each iteration; the number of training epochs represents the complete number of times the model is trained on the entire training data; the batch size defines the number of samples used to update the weights each time. Specifically, the training can adopt the Adam optimizer, a learning rate of 0.01, 150 training epochs, and a batch size of 156.

[0176] Figure 10This article shows an example of the PewLSTM model architecture and its possible input and output dimensions. The model consists of an input layer, two hidden layers, and an output layer. The input layer has three dimensions: the first dimension corresponds to the batch size; the second dimension corresponds to the amount of historical distance data used for prediction (if the location is collected every 5 minutes between 8:00 and 21:00, the number of historical data points is 156); and the third dimension depends on the number of weather features. For example, when using temperature and humidity as weather features, the third dimension of the input vector is 3—the first feature is the distance vector, and the temperature and humidity vectors serve as the second and third features, respectively. Figure 10 The second and third layers in the architecture are hidden layers, responsible for extracting features from the input data and establishing associations between distance and core attendance periods. The final layer is the output layer, which generates the PewLSTM output after applying a sigmoid activation function and calculating a binary cross-entropy loss. For example, a 26-dimensional output vector can be used to predict core attendance periods in 30-minute intervals between 8:00 AM and 9:00 PM for the next three days.

[0177] Figure 11 The following schematic diagram provides an overview of one embodiment of the energy-saving service delivery method. A recipient carries a mobile device 1110 daily, with an installed application linked to a service delivery entity 1105 and a time prediction system 1115. During installation, the application, the mobile device, or the recipient is assigned an identification code. In step 1112, the application periodically transmits its location information and identification code to the time prediction system 1115.

[0178] exist Figure 11 In the illustrated embodiment, after the entity 1105 receives an order to deliver a package to a recipient, it can notify the recipient of the expected delivery information via an application on the mobile device 1110 at step 1120. This allows the service recipient to track 1125 the delivery status of the package.

[0179] After receiving an order to deliver a package to a recipient, entity 1105 sends a request 1130 to time prediction system 1115. This request includes an identification code, the package's delivery location, expressed as a delivery address or geographic location, and the earliest feasible delivery date and time for entity 1105. Through this request, entity 1105 requests time prediction system 1115 to provide the recipient's predicted core presence time at the delivery address.

[0180] Time prediction system 1115 uses the information in entity 1105's request 1130 and the continuously received recipient's mobile device location data 1112 to calculate the recipient's core presence period at the delivery address. In step 1135, the system sends the predicted time period to entity 1105, which is calculated from the entity's earliest feasible delivery time and typically covers a range of two to three days.

[0181] With the help of core period data, entity 1105 can plan a delivery 1140 solution for the recipient. This method can reduce ineffective delivery attempts and inefficient delivery appointments, thereby reducing energy consumption in the service delivery process.

[0182] Figure 12 Another embodiment of the energy-saving method for delivering services is shown. Figure 12 In it, the scenario where the recipient 1205 orders goods from the online store 1210. Most of the steps shown in this figure are the same as Figure 11 the same, with the same reference numerals and will not be described again.

[0183] The recipient 1205 needs to complete registration in the online store 1210. During the registration process, the store will collect the recipient's name and delivery address or geographical location information. In this embodiment, the store will generate an identification code for the recipient and transmit this code to the application on the recipient's mobile device 1110 through step 1215 (this code can also be generated by the mobile application). This enables the mobile device to regularly send the converted location data carrying the ID to the time prediction system 1115 at step 1112.

[0184] Figure 12 The ordering process of the embodiment starts with the recipient 1205 placing an order in the online store through step 1220. After receiving the order, the store will send a delivery order to the delivery agent 1105. This agent can be an independent organization or a department under the store. The delivery order must contain the necessary information for the agent to complete the delivery, including the recipient's name, delivery address or geographical location, and the information of the goods / parcels to be delivered and the pick-up point.

[0185] After receiving the order, the delivery agent 1105 can perform the subsequent steps that are the same as Figure 11 those described.

[0186] Although the present invention has been illustrated by several specific embodiments, those skilled in the art should understand that various changes and modifications can be made to the implementation without departing from the principles of the invention defined by the claims. The present invention can also be embodied in other forms without departing from its core features. The embodiments should be considered exemplary in all aspects rather than restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description. Any changes that meet the equivalent meaning and scope of the claims fall within the scope of protection of this patent.

Claims

1. A method for reducing the energy consumption of delivering a service to a service recipient (1205) at a delivery address (1) or geographical location, the method comprising the following steps: 1.1 The mobile device (22; 1110) of the recipient sends at least to a time prediction system (3; 1115): 1.1.1 An identification code (2) corresponding to the recipient (1205); and 1.1.2 Data characterizing the location (5) of the mobile device (22; 1110), 1.1.3 This sending is carried out at fixed time intervals; 1.2 The time prediction system (3; 1115) receives and stores the data characterizing the location (5), and stores it together with the identification code (2) and timestamp of the recipient (1205); 1.3 An entity (1105) receives an order (1230) to deliver the service to the recipient (1205) at the delivery address (1) or geographical location; 1.3.1 The order at least includes: 1.3.1.1 The delivery address (1) or geographical location; and 1.3.1.2 Data for identifying the recipient; 1.3.2 The entity (1105) further receives the identification code (2) of the recipient (1205), or calculates the identification code of the recipient based on the data for identifying the recipient; 1.4 After receiving the order (1230), the entity (1105) sends a request (S205; 1130) to the time prediction system (3; 1115), asking the entity to provide the predicted presence core period (7) of the recipient (1205) at the delivery address (1) or geographical location; 1.4.1 The request (S205; 1130) at least includes: 1.4.1.1 The delivery address (1) or geographical location, and 1.4.1.2 The identification code (2) corresponding to the recipient (1205); 1.5 After receiving the request (S205; 1130), the time prediction system (3; 1115) calculates (S201) the requested predicted presence core period (7) of the recipient at the delivery address (1) or geographical location by using the location (5) received from the mobile device and the delivery address (1) or geographical location; 1.5.1 Provide (8; 1135) the predicted presence core period (7) of the requested recipient (1205) to the entity; 1.5.2 Thereby support the entity (1105) in delivering the service to the recipient (1205) at the delivery address or geographical location during the predicted presence core period (7) of the recipient at the delivery address.

2. A method for reducing the energy consumption of delivering the service to a service recipient (1205) at a delivery address (1) or geographical location, the method comprising the following steps: 2.1 A time prediction system (3; 1115) receives (S102; 1112) from the mobile device (22; 1110) of the recipient at fixed time intervals at least: 2.1.1 An identification code (2) corresponding to the recipient, and 2.1.2 Data characterizing the position (5) of the mobile device (22; 1110); 2.2 The time prediction system stores 2.2.1 The data characterizing the position (5), together with 2.2.2 The identification code (2) of the recipient and 2.2.3 A timestamp; 2.3 The time prediction system (3; 1115) receives a request from an entity to provide the predicted presence core period (7) of the recipient at the delivery address (1) or geographical location; 2.4 The request (S205; 1130) includes at least: 2.4.1 The delivery address (1) or geographical location, and 2.4.2 The identification code (2) corresponding to the recipient (1205); 2.5 After receiving the request (S205; 1130), the time prediction system provides (8; 1135) the requested predicted presence core period (7) to the entity; 2.5.1 Thereby enabling the entity (1105) to deliver the service to the recipient at the delivery address (1) or geographical location during the predicted presence core period (7) of the recipient at the delivery address.

3. A time prediction system, comprising means for performing the method described in the preceding claim.

4. A computer program comprising instructions, wherein, When the program is executed by a computer of the time prediction system (3; 1115) according to claim 3, the instructions cause the computer to perform the method according to claim 2; or 4.1 A computer-readable medium, having stored thereon the computer program.

5. A computer program for a mobile device (22; 1110) of a service recipient (1205), the program causing the mobile device to send at least to a time prediction system (3; 1115) performing the method according to claim 2: 5.1 An identification code (2) corresponding to the recipient; and 5.2 Data characterizing the position (5) of the mobile device, 5.3 The sending is performed at fixed time intervals.

6. A method performed by an entity (1105) involved in delivering a service to a recipient (1205) at a delivery address (1) or geographical location, the method comprising the following steps: 6.1 The entity (1105) receives an order to deliver the service to the recipient (1205) at the delivery address (1) or geographical location. 6.1.1 The order (1230) includes at least: 6.1.1.1 The delivery address (1) or geographical location; and 6.1.1.2 The name of the recipient. 6.1.2 The entity further receives the identification code (2) corresponding to the recipient, or calculates the identification code (2) corresponding to the recipient based on the relevant information of the recipient. 6.2 After receiving the order (1230), the entity sends a request (S205; 1130) to the time prediction system (3; 1115) as claimed in claim 3 to receive from the time prediction system the predicted presence core period (7) of the recipient at the delivery address (1) or geographical location. 6.2.1 The request includes at least: 6.2.1.1 The delivery address (1) or geographical location; and 6.2.1.2 The identification code (2) corresponding to the recipient (1205). 6.3 After sending the request (S205; 1130), the entity receives (8; 1135) from the time prediction system the requested predicted presence core period (7) of the recipient. 6.3.1 Thereby enabling the entity (1105) to deliver the service to the recipient at the delivery address or geographical location during the predicted presence core period (7) of the recipient at the delivery address (1) or geographical location.

7. The method according to any one of claims 1 or 2, wherein 7.1 Before the data of the position (5) of the mobile device (22; 1110) representing the recipient (1205) is sent to the time prediction system (3; 1115), it is transformed on the mobile device of the recipient, 7.1.1 such that the time prediction system cannot locate the position of the mobile device on the earth, 7.1.2 but the transformed position data allows the time prediction system (3; 1115) to measure the distance between different positions of the mobile device within a predetermined distance range; 7.2 The delivery address (1) or geographical location for delivering the service is transformed in the same way; 7.3 Preferably, the transformation of the data representing the position of the mobile device adopts at least one of the following methods: 7.3.1 Perform geohashing on the position data and delete an appropriate number of leading digits in the geohash code, preferably delete the first three digits; and 7.3.2 Delete an appropriate number of leading digits in the geographical coordinates of the position, preferably delete the integer geographical degree value and retain the geographical minute and second value.

8. The method according to any one of the preceding method claims, wherein 8.1 The identification code (2) corresponding to the recipient (1205) is established in such a way that the time prediction system (3; 1115) cannot reconstruct the name of the recipient, the delivery address (1) or geographical location. 8.2 Preferably, the identification code (2) is generated from data identifying the recipient, and is generated in such a way that the data identifying the recipient (1205) cannot be reconstructed from the identification code, preferably by hashing the data identifying the recipient, more preferably by hashing the email address of the recipient.

9. The method according to any one of the preceding method claims, characterized in that 9.1 The group of mobile devices corresponding to the group of recipients sends to the time prediction system at least: 9.1.1 The identification code (2) corresponding to an individual recipient, or an identification code shared by all recipients from the group of recipients; and 9.1.2 Data characterizing the location of the mobile device of the individual recipient, 9.1.3 This sending is carried out at fixed time intervals; 9.2 Wherein the time prediction system calculates the predicted presence core period (7) of any recipient in the group of recipients at the delivery address (1) or geographical location.

10. The method according to any one of the preceding method claims, characterized in that 10.1 The time prediction system (3; 1115) calculates the probability that the mobile device (22; 1110) maintains a preset distance from the delivery address (1) or geographical location at a specified moment; and 10.2 By comparing the calculated probability with a preset threshold probability, the predicted presence core period (7) of the recipient (1205) at the delivery address (1) or geographical location is deduced.

11. The method according to the preceding claim, characterized in that Machine learning or deep learning methods are used to predict the presence core period (7), preferably using a recurrent neural network model, a long short-term memory (LSTM) model, such as a periodic LSTM with a meteorological perception gating mechanism or a Transformer-based model.

12. The method (100) according to claim 10 or 11, further comprising: 12.1 Providing additional information (15), preferably meteorological data, to the time prediction system (3; 1115); and 12.2 The time prediction system calculates the predicted presence core period (7) of the recipient (1205) at the delivery address (1) or geographical location according to the additional information.

13. The method according to any one of the preceding method claims, further comprising any one of the following steps: 13.1 Using a verification code to confirm whether the mobile device has the right to transmit location characterization data to the time prediction system; 13.2 Sending the location characterization data from the mobile device to the time prediction system through a proxy server; 13.3 Generating an event token for authorizing a request to the time prediction system for the predicted presence core period (7) of the recipient at the delivery address (1) or geographical location in this delivery event, and invalidating the event token after the service is successfully delivered to the recipient at the delivery address (1) or geographical location (S208).

14. The method according to any one of the preceding method claims, further comprising any one of the following steps: 14.1 Before delivering the service to the recipient at the delivery address or geographical location, verifying the actual presence status of the recipient (1205) at the delivery address (1) or geographical location, wherein the verification is completed by the entity (1105) sending a corresponding request to the time prediction system (3; 1115); 14.2 Monitoring, in the time prediction system, the distance between the converted location of the mobile device and the converted delivery address (1) or the converted geographical location; determining whether the mobile device is at the converted delivery address (1) or the converted geographical location, or whether it can reach the converted delivery address (1) or the converted geographical location in time during the predicted presence core period; If the mobile device cannot reach the converted delivery address (1) or the converted geographical location within the predicted presence core period, updating the predicted presence core period.

15. A compliance data augmentation processing method, comprising the following steps: 15.1 Collecting data at a collection point; 15.2 Anonymizing the data at the collection point; 15.3 Transmitting the anonymized data to a data room; 15.4 Performing an augmentation process on the anonymized data in the data room, 15.4.1 wherein personalized recommendations or insights are obtained from the data by augmenting the anonymized data; 15.5 Retrieving the personalized recommendations or insights using a cryptographic key, 15.5.1 wherein the possibility of retrieving the personalized recommendations or insights is restricted in time.

Images