Method for determining arrival time uncertainty and related methods and devices

Abstract

The present invention discloses a method for determining arrival time uncertainty and related methods and devices. The method includes: using historical planned roads and the actual uncertainties corresponding to the historical planned roads to train a preset uncertainty neural network model, and learning the weights in the uncertainty neural network model that affect the uncertainty; the actual uncertainty is the ratio of the absolute value of the difference between the actual travel time and the estimated arrival time (ETA) of the historical planned road to the actual travel time; splitting the planned road into at least one road section and at least one intersection, using the identifiers of the at least one road section and at least one intersection as inputs to the uncertainty neural network model, and obtaining the theoretical uncertainty of the planned road. The present invention achieves accurate calculation of the estimated arrival time uncertainty, making it easier for users to choose roads that are more likely to arrive on time, facilitating user travel, and improving the user experience of using navigation software.

Description

Technical Field

[0001] The present invention relates to the field of geographic information technology, and in particular to a method for determining arrival time uncertainty, a method for recommending planned roads, a method for determining punctuality, and related devices. Background Art

[0002] In the field of electronic maps, most electronic maps nowadays support navigation functions. The map navigation function will automatically plan multiple routes for users based on the destination entered by the user. Each route will correspond to a different estimated time of arrival (ETA). Users can reasonably plan their itinerary, travel routes, and travel methods based on the ETA provided by the map navigation function. However, it is inevitable that there will always be some error between the ETA and the user's actual travel time (the actual travel time of the user through a section of road). Therefore, the ETA provided by the map navigation function has different actual uncertainties. Among them, T real_travel_time is the actual running time, T eta When the above errors are small, the uncertainty of ETA is low; when the above errors are large, the uncertainty of ETA is high.

[0003] Most map navigation apps on the market don't offer ETA (Earnings Time Added) accuracy, preventing users from choosing routes with the most accurate ETAs. Choosing a route with the highest ETA uncertainty can lead to missed arrival times and potentially significant losses and unnecessary inconvenience.

[0004] ETA uncertainty is particularly important in certain situations. For example, a user needs to arrive at the airport at a certain time, and the map navigation function has planned two routes for the user. These two routes have different ETAs and different ETA uncertainties. In this case, the user is often more interested in knowing which route has the lowest ETA uncertainty. In other words, the user is more interested in knowing which route will have the highest probability of arriving at the airport within the ETA provided by the map navigation function without delaying the trip. This is where the value of ETA on-time performance becomes apparent.

[0005] In view of the above situation, it is necessary for map navigation software to provide users with ETA punctuality or ETA uncertainty functions so that users can arrive at their destination at a more accurate time and facilitate their travel. Summary of the Invention

[0006] In view of the above problems, the present invention is proposed to provide a method for determining arrival time uncertainty, a method for recommending planned roads, a method for determining punctuality and related devices that overcome the above problems or at least partially solve the above problems.

[0007] In a first aspect, an embodiment of the present invention provides a method for determining arrival time uncertainty, comprising:

[0008] Using historical planned roads and actual uncertainties corresponding to the historical planned roads, a preset uncertainty neural network model is trained to learn the weights in the uncertainty neural network model that affect the uncertainty; the actual uncertainty is the ratio of the absolute value of the difference between the actual travel time and the estimated arrival time (ETA) of the historical planned roads to the actual travel time;

[0009] The planned road is split into at least one road section and at least one intersection, and the identifiers of the at least one road section and the at least one intersection are used as inputs of an uncertainty neural network model to obtain a theoretical uncertainty of the planned road.

[0010] In one embodiment, the preset uncertainty neural network model is pre-created in the following manner:

[0011] Create a model that includes the sum of road uncertainty and intersection uncertainty; where:

[0012] The road uncertainty includes: road uncertainty in a congested state or road uncertainty in a non-congested state.

[0013] In one embodiment, the uncertainty of the road in the congested state includes: the uncertainty of the historical congestion of the road and / or the uncertainty of the current congestion of the road.

[0014] In one embodiment, the road attributes include any one or more of the following: travel time attributes, travel duration attributes, and road grade attributes;

[0015] The intersection attributes include any one or more of the following: an intersection type attribute and a turning direction attribute.

[0016] In a second aspect, an embodiment of the present invention provides a method for recommending planned roads, including:

[0017] For the obtained planned road, split the planned road into at least one road section and at least one intersection according to the connection relationship between the roads and the intersections;

[0018] Calculating a theoretical uncertainty of an arrival time corresponding to a planned road based on at least one road segment and at least one intersection and a preset model of uncertainty of an estimated arrival time;

[0019] Recommend the corresponding planned road to the user based on the calculated theoretical uncertainty of the planned route;

[0020] Among them, the theoretical uncertainty of the arrival time corresponding to the planned road is calculated based on at least one road section and at least one intersection, and a preset model of estimated arrival time uncertainty, and is determined by using the aforementioned method for determining the arrival time uncertainty.

[0021] In one embodiment, based on the calculated theoretical uncertainty of the planned route, the corresponding planned road is recommended to the user, including:

[0022] Sort the calculated planned roads by size and recommend to the user a set number of planned roads starting from the one with the smallest theoretical uncertainty; or

[0023] Recommend planned roads to users whose theoretical uncertainty is less than or equal to the preset threshold.

[0024] In a third aspect, an embodiment of the present invention provides a method for determining punctuality, including:

[0025] Determine the normal distribution X~N(0,σ 2 );

[0026] According to the preset punctuality rate interval and the determined normal distribution X~N(0, σ 2 ), determining a proportion of users falling within the punctuality rate range, and determining the user proportion as the punctuality rate of the road;

[0027] The road arrival uncertainty obeys the normal distribution X~N(0, σ 2 ) is determined by using the aforementioned method for determining the uncertainty of the arrival time.

[0028] In one embodiment, the normal distribution X~N(0, σ 2 The value of σ in ) is determined by the following formula: E(|x|) is the theoretical uncertainty determined by the above-mentioned method for determining the uncertainty of the arrival time.

[0029] In one embodiment, the method for determining the punctuality rate further includes:

[0030] When recommending a planned road, the punctuality rate of the planned road is also recommended to the user.

[0031] In a fourth aspect, an embodiment of the present invention provides a device for determining arrival time uncertainty, including:

[0032] A model training module is configured to train a preset uncertainty neural network model using historical planned roads and actual uncertainties corresponding to the historical planned roads, and learn the weights in the uncertainty neural network model that affect the uncertainty; the actual uncertainty is the ratio of the absolute value of the difference between the actual travel time and the estimated time of arrival (ETA) of the historical planned roads to the actual travel time;

[0033] A road splitting module is used to split the planned road into at least one road section and at least one intersection;

[0034] The uncertainty calculation module is used to use the identification of the at least one road section and the at least one intersection as the input of the uncertainty neural network model, calculate the theoretical uncertainty corresponding to the at least one road section and the at least one intersection and accumulate them to obtain the theoretical uncertainty of the planned road.

[0035] In a fifth aspect, an embodiment of the present invention provides a device for recommending planned roads, including:

[0036] A road splitting module is used to split the obtained planned road into at least one road section and at least one intersection according to the connection relationship between the roads and intersections;

[0037] an uncertainty calculation module, configured to calculate a theoretical uncertainty of an arrival time corresponding to a planned road based on at least one road segment and at least one intersection and a preset estimated arrival time uncertainty model;

[0038] The road recommendation module is used to recommend the corresponding planned roads to users based on the calculated theoretical uncertainty of the planned route;

[0039] Among them, the theoretical uncertainty of the arrival time corresponding to the planned road is calculated based on at least one road section and at least one intersection, and a preset model of estimated arrival time uncertainty, and is determined by using the aforementioned method for determining the arrival time uncertainty.

[0040] In a sixth aspect, an embodiment of the present invention provides a device for determining punctuality, including:

[0041] Distribution determination module, used to determine the normal distribution X~N(0,σ 2 );

[0042] The user ratio determination module is used to determine the user ratio according to the preset punctuality rate interval and the determined normal distribution X~N(0, σ 2 ), determining a proportion of users falling within the punctuality rate range, and determining the proportion of users as the punctuality rate;

[0043] The road arrival uncertainty obeys the normal distribution X~N(0, σ 2 ) is determined by using the method for determining the uncertainty of the arrival time as described above.

[0044] In the seventh aspect, an embodiment of the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implements the aforementioned method for determining the uncertainty of arrival time or the aforementioned method for recommending planned roads, or the aforementioned method for determining the punctuality rate.

[0045] In an eighth aspect, an embodiment of the present invention provides a navigation client, comprising: a memory and a processor; wherein the memory stores a computer program, which, when executed by the processor, can implement the aforementioned method for determining the uncertainty of arrival time or the aforementioned method for recommending planned roads, or the aforementioned method for determining the punctuality rate.

[0046] The beneficial effects of the above technical solutions provided by the embodiments of the present invention include at least:

[0047] 1. The above-mentioned method for determining arrival time uncertainty and the related method and device for recommending planned roads provided by the embodiments of the present invention learn the weights of various dimensions affecting uncertainty in the neural network model of ETA uncertainty based on historical planned roads and the actual uncertainties corresponding to the historical planned roads, thereby obtaining a calculation model that can accurately predict ETA uncertainty. After splitting the planned route into roads and intersections, calculations are performed according to the uncertainty model to obtain the corresponding uncertainty. Based on the size of the calculated uncertainty, planned roads with lower uncertainty are recommended to the user, thereby achieving accurate calculation of the estimated arrival time uncertainty, facilitating users to choose roads that are more likely to arrive on time, facilitating users' travel, and improving users' user experience of using navigation software.

[0048] 2. The punctuality determination method and related apparatus provided in the embodiments of the present invention use ETA uncertainty to derive the ratio of the number of users in a corresponding interval centered on the ETA to the total number of users, thereby achieving an accurate estimation of the road's punctuality. This provides users with a reference for selecting planned routes from another perspective, making it easier for users to independently select roads with higher punctuality rates. It can also provide a reference for navigation software manufacturers in their punctuality assessments, facilitating their provision of better navigation route recommendation services.

[0049] 3. The ETA uncertainty model provided by the embodiments of the present invention fully considers factors that may affect uncertainty in various situations, such as the impact of various road attributes on uncertainty in congested and non-congested situations, the impact of intersection type and turning direction on ETA uncertainty, and the impact of user driving behavior on uncertainty in non-congested situations. These factors are used to construct an uncertainty neural network model, allowing the model to cover as many multi-dimensional situations as possible, thereby being as close as possible to the uncertainty in real situations, thereby achieving accurate uncertainty prediction calculations. Since the on-time rate is also converted from the ETA uncertainty model, the calculation of the on-time rate can also reflect the impact of various factors on the on-time rate in multiple dimensions.

[0050] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0051] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0053] Figure 1 A flowchart of a method for determining arrival time uncertainty provided by an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of creating an uncertainty neural network model provided by an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of uncertainty prediction calculation using an uncertainty neural network model in an embodiment of the present invention;

[0056] Figure 4 A flowchart of a method for recommending planned roads provided in an embodiment of the present invention;

[0057] Figure 5 A flowchart of a method for determining punctuality provided by an embodiment of the present invention;

[0058] Figure 6 A schematic diagram of the structure of an apparatus for determining uncertainty of arrival time provided by an embodiment of the present invention;

[0059] Figure 7A schematic diagram of the structure of a device for recommending planned roads provided in an embodiment of the present invention;

[0060] Figure 8 A schematic diagram of the structure of an apparatus for determining punctuality provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0062] The inventors of the present invention have discovered that there are always some errors between the ETA and the user's actual travel time (the actual time it takes the user to travel through a section of road), and these errors are related to various factors.

[0063] For example: Peak travel

[0064] Peak travel: The routes planned by the system may be congested or non-congested, and there is a large deviation between the actual arrival times of users in these two situations.

[0065] Travel during off-peak hours: Traffic conditions are relatively stable during this time, and users are more likely to arrive within the planned ETA time.

[0066] Or, as the user's scenario changes, there will be some new problems:

[0067] Catching a flight at the airport: The system plans multiple routes with similar ETAs. At this time, users are more concerned about which route has a higher punctuality rate, that is, the user is more likely to arrive at the ETA time, otherwise the user may be delayed.

[0068] Going home for dinner after work: You only need to know what time you will arrive; you don’t need to strictly plan your arrival time according to the ETA.

[0069] The actual arrival time of a user is a series of distributions around the ETA. The shape of these distributions is related to many factors, such as the peak time of the day, the travel time, the complexity of the road, etc.

[0070] Therefore, it is necessary to establish an ETA uncertainty neural network model to measure the degree of deviation between the actual travel time and the estimated travel time for each planned road. The degree of deviation is equivalent to the expectation of the absolute value of the normal distribution. The larger the value, the higher the uncertainty of the ETA.

[0071] Based on the above-mentioned idea of solving the problem, the embodiment of the present invention provides a method for determining the uncertainty of the arrival time, referring to Figure 1As shown, the following steps are included:

[0072] S11. Using historical planned roads and actual uncertainties corresponding to the historical planned roads, a preset uncertainty neural network model is trained to learn weights in the uncertainty neural network model that affect the uncertainty; the actual uncertainty is the ratio of the absolute value of the difference between the actual travel time and the estimated time of arrival (ETA) for the historical planned roads to the actual travel time;

[0073] S12. Split the planned road into at least one road section and at least one intersection, and use the identifiers of the at least one road section and the at least one intersection as inputs to an uncertainty neural network model to obtain a theoretical uncertainty of the planned road.

[0074] It should be noted that the theoretical uncertainty mentioned in the embodiments of the present invention is calculated using an uncertainty neural network model, which is a value close to the true uncertainty, but not the actual uncertainty under real circumstances. Therefore, in the embodiments of the present invention, it is referred to as theoretical uncertainty to distinguish it from the true uncertainty.

[0075] In real life, there are many factors that affect the uncertainty of ETA. For example:

[0076] 1. Travel time: Travel time is divided into peak and off-peak hours. During peak hours, roads are prone to congestion, with traffic frequently fluctuating between congested and unobstructed conditions. Due to the limited ability of the ETA model to predict traffic conditions, traffic forecasts are subject to significant uncertainty. Late at night or in the early morning, roads are more likely to be unobstructed and less likely to experience frequent traffic changes, resulting in less uncertainty in traffic forecasts.

[0077] 2. Road Class: Different road classes introduce different levels of uncertainty. High-class roads have more regular traffic patterns, making them easier to predict. Low-class roads, however, face challenges due to potential unexpected events, such as school traffic and vehicle U-turns on narrow sections. These events can introduce changes in road conditions, making them less regular and more challenging to predict.

[0078] 3. Driving time: Since road condition predictions are based on historical and current road conditions, the further into the future, the less information is available. Therefore, the closer to the present, the more accurate the road condition predictions are, and the further into the future, the greater the deviation in the road condition predictions. In other words, the further into the future, the greater the uncertainty caused by the road condition predictions.

[0079] 4. Intersections: Uncertainty exhibits different characteristics due to the varying complexity of intersections and the user's steering actions. For more complex intersections, the phase relationship between traffic lights can lead to greater uncertainty. Complex steering maneuvers, such as U-turns and left turns, also introduce significant uncertainty.

[0080] Based on the above factors, Figure 2 As shown, in one embodiment, the process of creating the uncertainty neural network model can be achieved by creating a model that includes the sum of road uncertainty and intersection uncertainty;

[0081] The road uncertainty may include the road uncertainty in a congested state or the road uncertainty in a non-congested state.

[0082] The uncertainty neural network model provided by the embodiment of the present invention is composed of road uncertainty and intersection uncertainty. The road uncertainty can include road uncertainty in a congested state and road uncertainty in a non-congested state depending on whether congestion occurs.

[0083] In congested conditions, uncertainty primarily stems from the uncertainty of road condition predictions. Congestion can be further categorized as SP congestion and AutoLR congestion. The former represents historical congestion on the current road, while the latter represents real-time congestion on the current road. Therefore, in this uncertainty neural network model, road uncertainty in congested conditions can include uncertainty about historical congestion and / or uncertainty about current congestion.

[0084] In non-congested conditions, driving behavior is less affected by road condition predictions, and the primary source of uncertainty is the user's personalized driving preferences. In this case, uncertainty can be defined as the sum of the user's personalized attributes and ETA, or the uncertainty introduced by the user's free flow.

[0085] Specifically, based on the above-mentioned different road conditions and intersection conditions, the arrival time uncertainty model provided by the embodiment of the present invention can be expressed by the following formula:

[0086]

[0087] In the above formula:

[0088] If_autolr_congest and If_sp_congest are the values for judging whether there is historical congestion and current congestion respectively. If there is congestion, it is 1, and if there is no congestion, it is 0;

[0089] X linkis the discretized road attribute; the road attribute includes one or more of the following: travel time attribute (peak time), travel time attribute, and road grade attribute;

[0090] Peak hours: 7:00-8:00, 8:00-9:00, 9:00-10:00...

[0091] Driving time: 0min-5min, 5min-10min, 10min-15min...

[0092] Road levels: highway, expressway, main road, secondary road...

[0093] After the above attributes are discretized, the corresponding attribute value is obtained, namely X link The value of

[0094] W autolr is the weight corresponding to the discretized road attribute when the current congestion occurs; where W autolr =concat(W autolr_peak_hour ,W autolr_period ,W autolr_road_class ), which contains three sets of attribute values (peak time attribute, travel time attribute, road grade attribute), and concat is used to connect two or more arrays.

[0095] W sp is the weight corresponding to the discretized road attributes when historical congestion occurs; where W sp =concat(W sp_peak_hour ,W sp_period ,W sp_road_class ), which also contains three sets of attribute values (peak hour attribute, travel time attribute, and road grade attribute);

[0096] T autolr is the actual travel time of the road, which can be obtained directly;

[0097] T sp is the historical average travel time of the road; it can be obtained directly;

[0098] W freeflow It is the weight corresponding to the user behavior attribute when there is no historical congestion or current congestion; that is, the user's free flow weight.

[0099] X turnis the discretized intersection attribute; intersection attributes may include intersection type and turning action attributes; intersection types include: intersection without traffic lights, intersection with traffic lights but no fork, simple intersection with traffic lights, complex intersection with traffic lights, etc. Turning actions include: going straight, turning left, turning right, U-turn, etc.; these types are discretized to obtain corresponding attribute values;

[0100] W turn is the weight corresponding to the discretized intersection attribute.

[0101] In one embodiment, in the above step S11, specifically, the weights corresponding to the road attributes, intersection attributes, and user behavior attributes in the formula of the uncertainty neural network model can be obtained by training the neural network using the real uncertainty of the historical planned route as a sample. Specifically, the historical planned road is split into each road and each intersection, and the cumulative uncertainty value ΣT is calculated according to the uncertainty neural network model. pred , substitute into the following loss formula:

[0102]

[0103] In the above formula, T real_travle_time is the actual travel time of the road, T eta ETA is the estimated time of arrival;

[0104] Through the learning of neural network, the W in the uncertainty neural network model is continuously corrected in the direction of reducing the loss. autolr 、W sp 、W turn 、W freeflow The value of W in the uncertainty neural network model is determined until the loss is closest to the actual uncertainty of the historical planned road. autolr 、W sp 、W turn 、W freeflow The value of .

[0105] Based on the above ETA uncertainty neural network model, it can be deduced that the relationship between the factors affecting uncertainty and uncertainty, as well as the relationship between these factors are as follows:

[0106] Congestion, whether in SP or AutoLR congestion, depends on the impact of three factors on the uncertainty neural network model: travel time, trip time, and road grade. These three factors act on the uncertainty neural network model by multiplying the travel time under SP and AutoLR congestion, respectively. The relationship between these three factors and uncertainty is explained in detail below.

[0107] The driving time factor shows a trend of first decreasing and then increasing under SP congestion, approaching 0 at 40-45 minutes. This indicates that the driving time factor has almost no effect on the uncertainty neural network model during this period. After 45 minutes, the driving time factor begins to rise rapidly, indicating that after the driving time exceeds a certain range, the driving time factor has a significant impact on the ETA uncertainty neural network model.

[0108] In the case of AutoLR congestion, the driving time factor shows a downward trend. In the 0-5 minute time period, the driving time factor is closest to 0. As the driving time increases, the driving time factor continues to decrease, indicating that as time goes by, the impact of real-time road conditions on the ETA uncertainty neural network model gradually decreases.

[0109] The travel time factor can be divided into 24 time periods according to the number of whole hours. Generally speaking, the morning and evening peaks refer to 7:00-8:00 and 17:00-19:00 respectively. During these two time periods, the travel time factor will reach its peak accordingly. Whether in the case of SP congestion or AutoLR congestion, the change trend of the travel time factor is the same, but the travel time factor of AutoLR congestion is always larger than that of SP congestion. A speculative explanation can be made here. People's working hours are relatively scattered. After the unified morning peak ends, many people still have their own travel needs. This personal travel demand is still highly correlated with the real-time road conditions.

[0110] Road grade factors. Because historical road conditions (SP) reflect traffic regularity, lower road grades indicate less regular traffic, making road condition prediction more difficult and significantly impacting the ETA uncertainty neural network model. Real-time traffic conditions (AutoLR) reflect the sporadic nature of traffic. Since AutoLR only reports current, real-time traffic conditions, they are not significantly different across road grades.

[0111] When road uncertainty is not congested, the user's driving behavior factors into the uncertainty. This influence on the ETA uncertainty neural network model is primarily determined by the product of the user's degree of freedom parameter and the ETA. In a non-congested state, the degree of freedom parameter learned by the neural network is ɑ. In other words, in a smooth traffic flow, the time deviation caused by the user's behavior can be, for example, ɑ*ETA. For example, in a non-congested state, if the ETA is 30 minutes, the deviation from the actual travel time is approximately 2 minutes.

[0112] The arrival time uncertainty model provided by the embodiment of the present invention includes not only the road uncertainty mentioned above, but also the intersection uncertainty, wherein the factors affecting the intersection uncertainty include the turning direction factor and the intersection type factor.

[0113] Turning direction factors can be categorized into straight ahead, left turn, right turn, and U-turn. U-turns have the greatest impact on the ETA uncertainty neural network model, followed by left turn, right turn, and straight ahead. Intersection turning factors can be further categorized into intersections without traffic lights, simple intersections with traffic lights, and complex intersections with traffic lights. Complex intersections with traffic lights have the greatest impact on the ETA uncertainty neural network model, followed by intersections without traffic lights and simple intersections with green lights. This is consistent with everyday experience: ETA uncertainty is greater for complex road conditions than for simple road conditions. The greater the proportion of straight-ahead traffic on a road, the smaller the ETA uncertainty.

[0114] Problems with traffic prediction arise during peak hours, when congestion-related factors play a major role, while the user's driving preference (free flow) has little impact on uncertainty. During other times, the influence of the user's driving preference (free flow) on uncertainty increases.

[0115] The more straight lines there are on the road, the smaller the effect of steering on uncertainty.

[0116] After establishing the above-mentioned ETA uncertainty neural network model, the planned road can be divided into at least one road section and at least one intersection in the above-mentioned step S12, and the identification of the at least one road section and at least one intersection is used as the input of the uncertainty neural network model to obtain the theoretical uncertainty of the planned road.

[0117] Reference Figure 3 In the schematic diagram shown, the far left side represents a historical planned road provided by the navigation system, which is divided into several sections and intersections in the order of driving direction. The historical planned road is input as a sample into the uncertainty neural network model, and the uncertainty neural network model is trained to learn the weights of various factors that may affect the uncertainty in the neural network model. Then, the uncertainty of the planned road given to the user is predicted through the learned uncertainty neural network model.

[0118] The ETA uncertainty model provided by the embodiment of the present invention fully considers factors that may affect uncertainty in various situations, such as the impact of various road attributes in congested and non-congested situations on uncertainty, and the impact of user driving behavior on uncertainty in non-congested situations. Based on these factors, an uncertainty neural network model is constructed, so that the model can cover as many multi-dimensional situations as possible, so that it can be as close as possible to the actual uncertainty, thereby realizing accurate uncertainty prediction calculation.

[0119] The embodiment of the present invention also provides a method for recommending planned roads, referring to Figure 4 Shown, including:

[0120] S41. Split the obtained planned road into at least one road section and at least one intersection according to the connection relationship between the roads and intersections;

[0121] S42. Calculating a theoretical uncertainty of an arrival time corresponding to a planned road based on at least one road section and at least one intersection and a preset estimated arrival time uncertainty model;

[0122] S43. Recommending a corresponding planned road to the user based on the calculated theoretical uncertainty of the planned route.

[0123] In the above step S42, the theoretical uncertainty of the arrival time corresponding to the planned road is calculated based on the conditions of each road section and intersection, and the preset model of the uncertainty of the estimated arrival time, based on at least one road section and at least one intersection, and the preset model of the uncertainty of the estimated arrival time. The method for determining the uncertainty of the arrival time provided in the aforementioned embodiment of the present invention is used to determine it.

[0124] In one embodiment, in step S43, the corresponding planned roads are recommended to the user based on the calculated theoretical uncertainty of each planned route. This can be achieved in the following two ways:

[0125] The first method is to sort the calculated planned roads by size and recommend a set number of planned roads starting from the one with the smallest theoretical uncertainty to the user.

[0126] For example, the system generates several planned roads, calculates the corresponding uncertainties through the aforementioned ETA uncertainty neural network model, and then sorts them from small to large according to the uncertainty. The top three planned roads are recommended to the user, and the uncertainty value of each planned road is prompted.

[0127] The second method is to recommend planned roads to users whose theoretical uncertainty is less than or equal to a preset threshold.

[0128] The above-mentioned method for determining the uncertainty of arrival time and the related method for recommending planned roads provided by an embodiment of the present invention learn the weights of various dimensions affecting the uncertainty in the ETA uncertainty model based on historical planned roads and the actual uncertainties corresponding to the historical planned roads, thereby obtaining a calculation model that can accurately predict the ETA uncertainty. After splitting the planned route into roads and intersections, calculations are performed according to the uncertainty model to obtain the corresponding uncertainty. Based on the size of the calculated uncertainty, planned roads with lower uncertainty are recommended to users, thereby achieving accurate calculation of the uncertainty of the estimated arrival time, facilitating users to choose roads that are more likely to arrive on time, facilitating users' travel, and improving users' user experience of using navigation software.

[0129] The following describes a method for determining punctuality provided by an embodiment of the present invention.

[0130] The method for determining punctuality provided in the embodiment of the present invention is described in detail in Figure 5 As shown, the following steps are included:

[0131] S51. Determine the normal distribution X~N(0, σ 2 );

[0132] S52, according to the preset punctuality rate interval and the determined normal distribution X~N(0,σ 2 ), determine the proportion of users falling within the punctuality rate interval, and determine the user proportion as the punctuality rate of the road.

[0133] In the above step S51, the road arrival uncertainty obeys the normal distribution X~N(0, σ 2 ) can be determined by adopting the aforementioned method for determining the uncertainty of the arrival time provided by the embodiment of the present invention.

[0134] In the embodiment of the present invention, ETA punctuality is a concept relative to ETA uncertainty, which refers to the ratio of the number of users in an interval centered on ETA to the total number of users. In other words, it is the ratio of the number of users who arrive at the destination within the specified interval to the total number of users.

[0135] The uncertainty obtained by the uncertainty neural network model in the above method is actually the expectation of the absolute value of the actual travel time relative to the ETA distribution. Therefore, the distribution variance σ of this distribution can be converted through the normal distribution.

[0136] The conversion process is as follows:

[0137] Assuming a standard normal distribution X~N(0,1), the expected value calculation process of the normal distribution is as follows:

[0138]

[0139] make

[0140]

[0141] The following gamma function (Gamma function), also known as Euler's second integral, is used in the above calculation process. It is a type of function that extends the factorial function to real and complex numbers:

[0142]

[0143] Let t = k 2

[0144]

[0145] Further considering the case of non-standard normal distribution, E(|x|) is the uncertainty obtained by the uncertainty neural network model. Substituting it into the above formula, the value of σ can be deduced.

[0146] The reason why the above method is used to calculate the value of σ of the normal distribution instead of directly using the statistical variance method of historical data, such as the statistical value of σ obtained under the influence of different factors, is based on the following considerations:

[0147] 1) The data has many dimensions, such as peak hour, road type, intersection type, etc. The peak hour attribute alone has nearly 20 categories, and it is impossible to collect statistics for all of them.

[0148] 2) If there is more or less correlation between attributes, they cannot be simply superimposed according to independent distribution.

[0149] 3) Real data does not follow a standard normal distribution; its mean deviates slightly from the median, and there is a long tail on one side of the distribution. Because of this bias, ignoring the bias and performing direct calculations will result in significant errors when evaluating the final distribution.

[0150] In this embodiment of the present invention, a machine learning algorithm is used to determine uncertainty (the expectation of absolute values) across multiple dimensions. This uncertainty encompasses the characteristics of each dimension. For example, the uncertainty during peak hours is greater than at other times, and the uncertainty at complex intersections is greater than that at simple intersections. The σ calculated from this uncertainty also encompasses the characteristics of these attributes.

[0151] In other words, the learned on-time performance model should be stable across all attributes. For example, a road's on-time performance during peak hours and off-peak hours is roughly the same. The uncertainty calculated using the on-time performance model is the absolute value of the deviation of the actual travel time from the ETA. This uncertainty is actually the expectation factor that integrates all feature dimensions.

[0152] Taking the standard normal distribution X~N(0,1) obeyed by the road arrival uncertainty as an example, the interval of the punctuality rate is According to the calculation of the distribution function of the standard normal distribution, the probability of falling within this interval is: The table shows that the corresponding user ratio is approximately 57.6%. The calculation process is as follows:

[0153] For any interval The probability of falling within this interval is calculated as follows: In the case of standard normal distribution, let μ=0, σ=1. Then, where φ(x) is the distribution function of the standard normal distribution.

[0154] The embodiment of the present invention can also be based on the above normal distribution function X~N(0, σ 2 ), and using the known user ratios, we can infer the on-time rate ranges that these user ratios fall into. The specific calculation process is the reverse of the above process and will not be repeated here.

[0155] The aforementioned ETA uncertainty model provided in the embodiment of the present invention fully considers the factors that may affect the uncertainty in various situations, so that the model can cover as many multi-dimensional situations as possible. Since the punctuality rate is also converted through the ETA uncertainty model, the calculation of the punctuality rate can also reflect the impact of various factors on the punctuality rate in multiple dimensions.

[0156] Based on the same inventive concept, an embodiment of the present invention also provides a device for determining arrival time uncertainty, a device for recommending planned roads, a device for determining punctuality, and related computer-readable storage media and a navigation client. Since the principles of the problems solved by these devices and navigation clients are similar to the aforementioned methods for determining arrival time uncertainty, determining punctuality, and recommending planned roads, the implementation of these devices and navigation clients can refer to the implementation of the aforementioned methods, and the repeated parts will not be repeated.

[0157] An embodiment of the present invention provides a device for determining the uncertainty of arrival time, referring to Figure 6 Shown, including:

[0158] A model training module 61 is configured to train a preset uncertainty neural network model using historical planned roads and actual uncertainties corresponding to the historical planned roads, and learn weights in the uncertainty neural network model that affect the uncertainty; the actual uncertainty is the ratio of the absolute value of the difference between the actual travel time and the estimated time of arrival (ETA) for the historical planned roads to the actual travel time;

[0159] A road splitting module 62 is used to split the planned road into at least one road segment and at least one intersection;

[0160] The uncertainty calculation module 63 is used to use the identification of the at least one road section and the at least one intersection as the input of the uncertainty neural network model, calculate the theoretical uncertainty corresponding to the at least one road section and the at least one intersection and accumulate them to obtain the theoretical uncertainty of the planned road.

[0161] In one embodiment, the above-mentioned device for determining the arrival time uncertainty also includes: a model creation module 64, which is specifically used to create a model including the sum of road uncertainty and intersection uncertainty; wherein: the road uncertainty includes: road uncertainty in a congested state or road uncertainty in a non-congested state.

[0162] In one embodiment, in the model creation module 64 , the uncertainty of the road in the congested state includes: the uncertainty of the historical congestion of the road and / or the uncertainty of the current congestion of the road.

[0163] In one embodiment, in the model creation module 64, the uncertainty of the road historical congestion is calculated by the following formula: If_autolr_congest*Reduce_sum(W autolr *X link )*T autolr ;

[0164] The uncertainty of the current congestion of the road is obtained by the following formula: If_sp_congest*Reduce_sum(W sp *X link )*T sp ;in:

[0165] The If_autolr_congest and If_sp_congest are respectively the values for judging whether there is historical congestion and current congestion, where the value is 1 if there is congestion and 0 if there is no congestion;

[0166] The X link is the discretized road attribute;

[0167] The W autolris the weight corresponding to the discretized road attributes when the current congestion occurs;

[0168] The W sp is the weight corresponding to the discretized road attributes when historical congestion occurs;

[0169] The T autolr is the actual travel time of the road;

[0170] The T sp is the historical average travel time of the road.

[0171] In one embodiment, in the above-mentioned model creation module 64, the uncertainty in the non-congested road state is obtained by the following formula: freeflow *X link ;

[0172] The W freeflow The weight corresponding to the user behavior attribute when there is no historical congestion or current congestion;

[0173] In one embodiment, in the model creation module 64, the uncertainty of the intersection is obtained by the following formula: turn *X turn ;

[0174] The X link is the discretized intersection attribute;

[0175] The W turn is the weight corresponding to the discretized intersection attribute.

[0176] In one embodiment, the road splitting module 62 is specifically used to split the historical planned roads into each road and each intersection, and calculate the cumulative uncertainty value ΣT according to the uncertainty neural network model. pred , substitute into the following loss formula:

[0177] In the above formula, T real_travle_time is the actual travel time of the road, T eta is the estimated time of arrival ETA; through the learning of the neural network, the W in the uncertainty neural network model is continuously corrected in the direction of reducing the Loss. autolr 、W sp 、W turn 、W freeflow The value of W in the uncertainty neural network model is determined until the loss is closest to the actual uncertainty of the historical planned road. autolr 、W sp 、W turn 、W freeflow The value of .

[0178] The embodiment of the present invention also provides a device for recommending planned roads, referring to Figure 7 Shown, including:

[0179] A road splitting module 71 is configured to split the obtained planned road into at least one road segment and at least one intersection according to the connection relationship between the roads and intersections;

[0180] an uncertainty calculation module 72 for calculating a theoretical uncertainty of an arrival time corresponding to a planned road based on at least one road segment and at least one intersection and a preset estimated arrival time uncertainty model;

[0181] A road recommendation module 73 is used to recommend a corresponding planned road to the user based on the calculated theoretical uncertainty of the planned route;

[0182] Among them, the theoretical uncertainty of the arrival time corresponding to the planned road is calculated based on at least one road section and at least one intersection, and a preset model of estimated arrival time uncertainty, and is determined by using the aforementioned method for determining the arrival time uncertainty.

[0183] The embodiment of the present invention also provides a device for determining punctuality, referring to Figure 8 Shown, including:

[0184] The distribution determination module 81 is used to determine the normal distribution X~N(0, σ 2 );

[0185] The user ratio determination module 82 is used to determine the user ratio according to the preset punctuality interval and the determined normal distribution X~N(0, σ 2 ), determining a proportion of users falling within the punctuality rate range, and determining the proportion of users as the punctuality rate;

[0186] The road arrival uncertainty obeys the normal distribution X~N(0, σ 2 ) is determined by using the aforementioned method for determining the uncertainty of the arrival time.

[0187] In one embodiment, the distribution determination module 81 specifically determines the normal distribution X~N(0, σ 2 ) in the value of σ: Where E(|x|) is the theoretical uncertainty determined by the method for determining the uncertainty of the arrival time.

[0188] An embodiment of the present invention also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implements the aforementioned method for determining the uncertainty of the arrival time or the aforementioned method for recommending a planned road, or the aforementioned method for determining the punctuality rate.

[0189] An embodiment of the present invention also provides a navigation client, comprising: a memory and a processor; wherein the memory stores a computer program, and when the program is executed by the processor, it can implement the aforementioned method for determining the uncertainty of the arrival time or the aforementioned method for recommending a planned road, or the aforementioned method for determining the punctuality rate.

[0190] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0191] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0192] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0193] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0194] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for determining the uncertainty of arrival time, characterized in that: include: Using historical planned roads and actual uncertainties corresponding to the historical planned roads, a preset uncertainty neural network model is trained to learn weights in the uncertainty neural network model that affect the uncertainty; The actual uncertainty is the ratio of the absolute value of the difference between the actual travel time and the estimated arrival time ETA of the historically planned road to the actual travel time; The planned road is split into at least one road section and at least one intersection, and the identifiers of the at least one road section and the at least one intersection are used as inputs of an uncertainty neural network model to obtain a theoretical uncertainty of the planned road.

2. The method according to claim 1, wherein The preset uncertainty neural network model is pre-created in the following way: Create a model that includes the sum of road uncertainty and intersection uncertainty; where: The road uncertainty includes: road uncertainty in a congested state or road uncertainty in a non-congested state.

3. The method according to claim 2, wherein The uncertainty of the road in the congested state includes: the uncertainty of the historical congestion of the road and / or the uncertainty of the current congestion of the road.

4. The method according to claim 1, wherein Using historical planned roads and actual uncertainties corresponding to the historical planned roads, a preset uncertainty neural network model is trained to learn the weights affecting the uncertainty in the uncertainty neural network model, including: The actual uncertainty of each road and intersection obtained by splitting the historical planned roads is used as the sample for neural network training. The neural network is trained and the W in the uncertainty neural network model is continuously corrected in the direction of reducing the preset loss. autolr 、W sp 、W turn and W freeflow The value of W in the uncertainty neural network model is determined until the loss is closest to the actual uncertainty of the historical planned road. autolr 、W sp 、W turn and W freeflow ; where: The W autolr is the weight corresponding to the discretized road attributes when the current congestion occurs; The W sp is the weight corresponding to the discretized road attributes when historical congestion occurs; The W turn is the weight corresponding to the discretized intersection attribute; The W freeflow It is the weight corresponding to the user behavior attribute when there is no historical congestion or current congestion.

5. The method according to claim 4, wherein The road attributes include any one or more of the following: travel time attributes, travel duration attributes, and road grade attributes; The intersection attributes include any one or more of the following: an intersection type attribute and a turning direction attribute.

6. A method for recommending planned roads, characterized in that: include: For the obtained planned road, split the planned road into at least one road section and at least one intersection according to the connection relationship between the roads and the intersections; Calculating a theoretical uncertainty of an arrival time corresponding to a planned road based on at least one road segment and at least one intersection and a preset model of uncertainty of an estimated arrival time; Recommend the corresponding planned road to the user based on the calculated theoretical uncertainty of the planned route; Wherein, the theoretical uncertainty of the arrival time corresponding to the planned road is calculated based on at least one road section and at least one intersection, and a preset estimated arrival time uncertainty model, and is determined by the method for determining the arrival time uncertainty as described in any one of claims 1-5.

7. The method according to claim 6, wherein Based on the calculated theoretical uncertainty of the planned route, the corresponding planned roads are recommended to the user, including: Sort the calculated planned roads by size and recommend to the user a set number of planned roads starting from the one with the smallest theoretical uncertainty; or Recommend planned roads to users whose theoretical uncertainty is less than or equal to the preset threshold.

8. A method for determining punctuality, characterized in that: include: Determine the normal distribution X~N(0,σ 2 ); According to the preset punctuality rate interval and the determined normal distribution X~N(0, σ 2 ), determining a proportion of users falling within the punctuality rate range, and determining the user proportion as the punctuality rate of the road; The road arrival uncertainty obeys the normal distribution X~N(0, σ 2 ) is determined by adopting the method for determining the uncertainty of the arrival time as described in any one of claims 1 to 5.

9. The method according to claim 8, wherein The method further comprises: When recommending a planned road, the punctuality rate of the planned road is also recommended to the user.

10. A device for determining the uncertainty of arrival time, characterized in that: include: A model training module is used to train a preset uncertainty neural network model using historical planned roads and actual uncertainties corresponding to the historical planned roads, and learn various weights in the uncertainty neural network model that affect the uncertainty; The actual uncertainty is the ratio of the absolute value of the difference between the actual travel time and the estimated arrival time ETA of the historically planned road to the actual travel time; A road splitting module is used to split the planned road into at least one road section and at least one intersection; The uncertainty calculation module is used to use the identification of the at least one road section and the at least one intersection as the input of the uncertainty neural network model, calculate the theoretical uncertainty corresponding to the at least one road section and the at least one intersection and accumulate them to obtain the theoretical uncertainty of the planned road.

11. A device for recommending planned roads, characterized in that: include: A road splitting module is used to split the obtained planned road into at least one road section and at least one intersection according to the connection relationship between the roads and intersections; an uncertainty calculation module, configured to calculate a theoretical uncertainty of an arrival time corresponding to a planned road based on at least one road segment and at least one intersection and a preset estimated arrival time uncertainty model; The road recommendation module is used to recommend the corresponding planned roads to users based on the calculated theoretical uncertainty of the planned route; Wherein, the theoretical uncertainty of the arrival time corresponding to the planned road is calculated based on at least one road section and at least one intersection, and a preset estimated arrival time uncertainty model, and is determined by the method for determining the arrival time uncertainty as described in any one of claims 1-5.

12. A device for determining punctuality, characterized in that: include: Distribution determination module, used to determine the normal distribution X~N(0,σ 2 ); The user ratio determination module is used to determine the user ratio according to the preset punctuality rate interval and the determined normal distribution X~N(0, σ 2 ), determining a proportion of users falling within the punctuality rate range, and determining the proportion of users as the punctuality rate; The road arrival uncertainty obeys the normal distribution X~N(0, σ 2 ) is determined by adopting the method for determining the uncertainty of the arrival time as described in any one of claims 1 to 5.

13. A navigation client, characterized in that: include: A memory and a processor; wherein the memory stores a computer program, which, when executed by the processor, can implement the method for determining the uncertainty of the arrival time as claimed in any one of claims 1 to 5, or the method for recommending a planned road as claimed in claim 6 or 7, or the method for determining the punctuality rate as claimed in claim 8 or 9.

Images