A self-service bus management method based on bus digital transformation

Abstract

The present invention discloses a self-service bus management method based on the digital transformation of public transportation, including: S1, providing an interactive interface to a user through a mobile phone terminal for the user to input a first interest vocabulary to be selected, wherein the first interest vocabulary includes: travel destination and travel purpose; S2, semantic association is performed based on the first interest vocabulary, the bus service type that the user may be interested in is obtained, and a second interest vocabulary is generated; S3, after the user selects the second interest vocabulary, the current location information and travel destination of the user's mobile phone terminal are extracted, and a number of recommended itinerary route information is generated according to the second interest vocabulary; S4, after the user determines the itinerary route information, the user's itinerary is navigated until the user arrives at the set travel destination; if the user re-enters the first interest vocabulary to be selected, the navigation is stopped. The present invention can provide diversified services according to the needs of passengers, saving user time in many aspects.

Description

Technical Field

[0001] The present invention relates to the field of public transportation digitalization, and in particular to a public transportation self-service management method based on public transportation digital transformation. Background Art

[0002] Public transportation has become an important part of people's travel mode, but the existing public transportation route selection is basically based on the travel route planning information formed by the departure point and the expected arrival point in the following way, including, for example, mechanized estimation based on the shortest time, the least transfer or the least walking. Although the way to realize public transportation transfer also uses mobile terminals as a carrier to realize mobile query, its method is relatively simple and cannot meet the personalized self-service requirements of users. Summary of the invention

[0003] The present invention provides a public transportation self-service management method based on public transportation digital transformation to overcome the above problems.

[0004] The present invention comprises:

[0005] S1, providing an interactive interface to a user through a mobile phone terminal for the user to input a first interest word to be selected, wherein the first interest word includes: travel destination, travel purpose;

[0006] S2, performing semantic association based on the first interest word, obtaining the bus service type that the user may be interested in, and generating a second interest word;

[0007] S3, after the user selects the second interest word, extract the current location information and travel destination of the user's mobile terminal, and generate a number of recommended itinerary route information according to the second interest word;

[0008] S4. After the user confirms the itinerary route information, the user's itinerary is navigated until the user reaches the set travel destination; if the user re-enters the first interest word to be selected, the navigation is stopped.

[0009] Furthermore, the input method of the first interest word in S1 includes: text input method and voice recognition input method; the voice recognition input method includes: collecting user audio data, performing audio analysis on the audio data to obtain corresponding voice text information and voice text word segmentation data; the calculation formula of audio analysis is:

[0010] fa=f(l n (f o (o),f(p))) (1)

[0011] Among them, o represents user audio data, fa represents the word segmentation data of speech text, p represents the standard Chinese pronunciation library, l n (*) indicates the features extracted by the twin network trained with contrast loss. The matching pairs of pronunciation and text are used to train the twin network. The features extracted by the twin network give the corresponding text according to the audio. f(*) indicates the semantic segmentation of text using the Bert network. f o (*) indicates that the audio is denoised.

[0012] Furthermore, in S2, semantic association is performed based on the first interest vocabulary, including: pre-matching a number of semantic association lists, wherein the semantic association lists are based on the travel destination as a keyword, and the types of public transportation services that can be provided by the destination are associated; the types of public transportation services include: transfer reminder services, travel services, catering services, accommodation services and shopping services.

[0013] Furthermore, performing semantic association based on the first interest vocabulary in S2 further includes: performing semantic association on the interest vocabulary based on a fuzzy matching algorithm, and if the interest vocabulary has words that cannot be directly matched, obtaining the words with the highest matching degree therewith.

[0014] Furthermore, the generation process of the recommended travel route information in S3 includes: taking the second interest word as a reference dimension, the second interest word is the bus service type, and determining the corresponding travel route based on the path planning algorithm; the calculation formula of the travel route is:

[0015] s=fusion(l(f(v),concat(f(x),fa)),m(f(y,z))) (2)

[0016] Among them, x represents the second interest word, y represents the departure location, z represents the destination, v represents the semantic association list, fa represents the word segmentation data of the speech text, m(*) represents the feature extraction of the text using the twin network, l(*) represents the feature extraction of the text using the convolutional network trained by cross entropy; f(*) represents the semantic segmentation of the text using the Bert network, concat represents the merging of the text features after semantic segmentation and the word segmentation data of the speech text into a one-dimensional feature, fusion(*) represents the fusion of the travel routes predicted by the two networks, and s represents the travel route;

[0017] Furthermore, the navigation of the user in S4 includes: obtaining in real time the distance between the user's current location and the nearest bus stop on the route to be taken, calculating the spatial distance between the passenger and the stop, and calculating the passenger's arrival time and the arrival time of the bus on the route to be taken.

[0018] The present invention can provide diversified services according to the needs of passengers, such as: providing transfer reminder services to prevent passengers from wasting time and energy due to missing their stops; providing estimated arrival time services to enable passengers to better plan their time; providing catering, tourism, shopping, accommodation and other services, and selectively making recommendations according to passenger needs, so that passengers have a better travel experience and save passengers the time of searching by themselves. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0020] Figure 1 It is a flow chart of the present invention;

[0021] Figure 2 The following is a flow chart of an embodiment. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Figure 1 It is a flow chart of the present invention, as Figure 1 As shown, the method of this embodiment may include:

[0024] S1, providing an interactive interface to a user through a mobile phone terminal for the user to input a first interest word to be selected, wherein the first interest word includes: travel destination, travel purpose;

[0025] Specifically, when the user only enters the travel destination: "Xinghai", a route to Xinghai will be planned based on the location of the user terminal. The planned routes include: routes with more restaurants, routes close to hotels, routes with more entertainment venues, etc.; if the user enters the travel destination: "Xinghai" and the travel purpose: "travel", routes to Xinghai, routes to attractions near Xinghai, and catering and accommodation services around Xinghai will be recommended.

[0026] S2, performing semantic association based on the first interest words, obtaining the types of public transportation services that the user may be interested in, and generating a second interest word;

[0027] Specifically, semantic association is performed according to the type of the first interest word. If the first interest word is a scenic spot and tourism, then tourism services, catering services, accommodation services, etc. near the scenic spot are associated.

[0028] S3, after the user selects the second interest word, extract the current location information and travel destination of the user's mobile terminal, and generate a number of recommended itinerary route information according to the second interest word;

[0029] Specifically, if the first interest word entered by the user is attractions and travel, and the second interest word is catering services, the recommended routes will include routes from the user's location to catering shops near the attractions, routes from the user's location to the attractions, routes from the user's location through catering shops near the attractions to the attractions, routes from the user's location to the attractions and then to catering shops near the attractions, etc.

[0030] S4. After the user confirms the itinerary route information, the user's itinerary is navigated until the user reaches the set travel destination; if the user re-enters the first interest word to be selected, the navigation is stopped.

[0031] Specifically, after the user selects a route, full navigation will be performed based on the real-time location of the user terminal. When the user location coincides with the destination, navigation will stop. If the user re-enters the first interest word on the way to the destination, navigation will stop, S1-S3 will be repeated, and then navigation will be performed to the new destination again.

[0032] Preferably, the input method of the first interest vocabulary in S1 includes: text input method and voice recognition input method; the voice recognition input method includes: collecting user audio data, performing audio analysis on the audio data to obtain corresponding voice text information and voice text word segmentation data; the calculation formula of audio analysis is:

[0033] f a =f(l n (f o (o),f(p))) (1)

[0034] Among them, o represents user audio data, fa represents the word segmentation data of speech text, p represents the standard Chinese pronunciation library, l n (*) indicates the features extracted by the twin network trained with contrast loss. The matching pairs of pronunciation and text are used to train the twin network. The features extracted by the twin network give the corresponding text according to the audio. f(*) indicates the semantic segmentation of text using the Bert network. f o(*) indicates that the audio is denoised.

[0035] Specifically, n (*) indicates the features extracted by the twin network trained with contrastive loss. The specific training of the twin network uses matching pairs of pronunciation and text. The twin network consists of two sub-networks. The input of the first sub-network is a pronunciation x1 in the user's audio data, which is then fed to the convolutional layer, pooling layer and fully connected layer in sequence, and finally outputs a feature vector f(x1); the final vector l(x1) is the encoding of the input x1, and the pronunciation x2 in the standard Chinese pronunciation library is input to the second sub-network, which is processed in the same way, and the encoding l(x2) of x(2) is obtained. Gradient descent is applied to l(x1) and l(x2) on the triple loss function to solve the distance between l(x1) and l(x2). If the distance is less than the threshold, it means that x1 and x2 are the same pronunciation, and then by finding the matching pair, The text corresponding to the pronunciation in the standard Chinese pronunciation library corresponding to the user audio can be obtained. f(*) indicates that the Bert network is used to perform semantic segmentation on the text. The Bert network training set includes the entire 3.2 billion texts. The input to Bert can be a paragraph of text or multiple paragraphs of text. Each paragraph of text is separated by "|||". The tagger actually tokenizes the sentence first. This step can separate spaces and periods, and then perform subword segmentation. The length of sentences in the data set is not necessarily equal. If it is long, it will be truncated, and if it is short, it will be filled. That is, it is determined whether the length of the entire sequence exceeds the requirement. If it exceeds, the last part of the excess will be truncated for a single sentence, and the end mark of the longer sentence will be deleted first for double sentences; if the two sentences are of equal length, the marks at the end of the two sentences will be deleted in turn until the total length reaches the requirement, f o (*) indicates that the audio is denoised.

[0036] Specifically, speech recognition is achieved by matching the user's pronunciation with text to obtain the recognition result; that is, the Twin Network denoises the user's speech, extracts audio with a similar pronunciation to the text, and recognizes the content in the audio through the text table entered in the Bert network, and finally obtains the recognition result.

[0037] Preferably, in S2, semantic association is performed based on the first interest vocabulary, including: pre-matching a number of semantic association lists, wherein the semantic association lists are based on the travel destination as a keyword, and the types of public transportation services that can be provided by the destination are associated; the types of public transportation services include: transfer reminder services, travel services, catering services, accommodation services and shopping services.

[0038] Specifically, it will provide attractions, shopping malls, and hotels near the destination and plan the corresponding travel routes. If a transfer is required during the route, a transfer reminder service will be provided based on the user's location during the navigation process.

[0039] Preferably, performing semantic association based on the first interest vocabulary in S2 further includes: performing semantic association on the interest vocabulary based on a fuzzy matching algorithm, and if the interest vocabulary has words that cannot be directly matched, obtaining the words with the highest matching degree therewith.

[0040] Specifically, the fuzzy matching algorithm associates the words of interest by matching the words input by the user with the words in the library, and selecting the word with the highest matching degree; Figure 2 As shown, if the user's first interest words are: "Xinghai" and "eating", the provided bus service must include the catering service with the highest matching degree with "eating".

[0041] Preferably, the process of generating the recommended itinerary route information in S3 includes: taking the second interest vocabulary as a reference dimension, the second interest vocabulary is the bus service type, and determining the corresponding travel itinerary route based on the path planning algorithm; the calculation formula of the travel itinerary route is:

[0042] s=fusion(l(f(v),concat(f(x),fa)),m(f(y,z))) (2)

[0043] Among them, x represents the second interest word, y represents the departure location, z represents the destination, v represents the semantic association list, fa represents the word segmentation data of the speech text, m(*) represents the feature extraction of the text using the twin network, l(*) represents the feature extraction of the text using the convolutional network trained by cross entropy; f(*) represents the semantic segmentation of the text using the Bert network, concat represents the merging of the text features after semantic segmentation and the word segmentation data of the speech text into a one-dimensional feature, fusion(*) represents the fusion of the travel routes predicted by the two networks, and s represents the travel route;

[0044] m(*) indicates that the features of the text are extracted using the twin network. When training the twin network, the matching pairs of features and travel routes are used. Therefore, extracting the features of the text means giving the predicted travel route based on the text. The twin network includes two sub-networks. The input of the first sub-network is a text x1, which is then fed to the convolutional layer, the pooling layer, and the fully connected layer in sequence, and outputs a feature vector l(x1); = vector l(x1) is the encoding of the input x1, and the corresponding travel route x2 is input to the second sub-network, which is processed in the same way, and the corresponding x(2) is obtained. The encoding l(x2) is obtained by applying gradient descent to l(x1) and l(x2) on the triple loss function to solve the distance between l(x1) and l(x2). If the distance is less than the threshold, it means that x1 and x2 are successfully matched. Then, by finding the matching pair, the predicted travel route can be given according to the text. l(*) indicates that the convolutional network trained with cross entropy is used to extract the features of the text. The matching pairs of features and travel routes are used when training the convolutional network. The input during training is a text x1 and a travel route x2, which are then fed to the convolutional layer, pooling layer, and fully connected layer in sequence. The convolutional network is ResNet-100. During the test, the input is a text, so the features of the text are extracted to give the predicted travel route according to the text. f(*) indicates that the Bert network is used to perform semantic segmentation on the text. The Bert network training set includes the entire 3.2 billion characters. The input to Bert can be a paragraph of text or multiple paragraphs of text. Each paragraph of text is separated by "|||". The tagger actually tokenizes the sentence first. This step can separate spaces and periods, and then perform subword segmentation. The length of sentences in the data set is not necessarily equal. If it is long, it will be truncated and if it is short, it will be filled. Specifically, it is to determine whether the length of the entire sequence exceeds the requirement. If it exceeds: the last part of the single sentence is cut off, and the end mark of the longer sentence is deleted first for the double sentence. If the two sentences are of equal length, the marks at the end of the two sentences are deleted in turn until the total length meets the requirement. concat indicates that the text features after semantic segmentation and the word segmentation data of the speech text are merged into a one-dimensional feature. fusion(*) indicates that the travel routes predicted by the two networks are weighted to obtain the final travel route s.

[0045] Specifically, a recommended travel route is calculated based on the first interest vocabulary, and then a travel route is recommended based on the second interest vocabulary, and finally the overlapping travel routes are selected as the final recommended travel route.

[0046] Preferably, navigating the user in S4 includes: obtaining in real time the distance between the user's current location and the nearest bus stop on the route to be taken, calculating the spatial distance between the passenger and the stop, and calculating the passenger's arrival time and the arrival time of the bus on the route to be taken.

[0047] Specifically, the distance to the nearest bus stop on the route the passenger is taking is retrieved through the user's current location, and the passenger's arrival time is determined based on the passenger's average walking speed; and the bus arrival time for the route is calculated based on the average speed of the bus closest to this stop and the distance between the bus the passenger is taking and the stop, allowing passengers to better plan their time.

[0048] Beneficial effects:

[0049] The present invention can provide diversified services according to the needs of passengers, such as: providing transfer reminder services to prevent passengers from wasting time and energy due to missing their stops; providing estimated arrival time services to enable passengers to better plan their time; providing catering, tourism, shopping, accommodation and other services, and selectively making recommendations according to passenger needs, so that passengers have a better travel experience and save passengers the time of searching by themselves.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A self-service bus management method based on bus digital transformation, characterized in that: include: S1, providing an interactive interface to a user through a mobile phone terminal for the user to input a first interest word to be selected, wherein the first interest word includes: travel destination, travel purpose; S2, performing semantic association based on the first interest word, obtaining the bus service type that the user may be interested in, and generating a second interest word; S3, after the user selects the second interest word, extract the current location information and travel destination of the user's mobile terminal, and generate a number of recommended itinerary route information according to the second interest word; S4, after the user confirms the itinerary route information, the user's itinerary is navigated until the user reaches the set travel destination; if the user re-enters the first interest word to be selected, the navigation is stopped; The generation process of the recommended itinerary route information in S3 includes: taking the second interest vocabulary as a reference dimension, the second interest vocabulary is the bus service type, and determining the corresponding travel itinerary route based on the path planning algorithm; the calculation formula of the travel itinerary route is: s=fusion(l(f(v),concat(f(x),fa)),m(f(y,z))) (2) Wherein, x represents the second interest word, y represents the departure location, z represents the destination, v represents the semantic association list, and fa represents the word segmentation data of the voice text. m(*) indicates the use of the twin network to extract the features of the text. When training the twin network, the matching pairs of features and travel routes are used. Therefore, extracting the features of the text means giving a predicted travel route based on the text. The twin network consists of two sub-networks. The input of the first sub-network is a text x1, which is then fed to the convolutional layer, pooling layer, and fully connected layer in sequence to output a feature vector l(x1); the vector l(x1) is the encoding of the input x1, and the corresponding travel route x2 is input to the second sub-network, which is processed in the same way, and the encoding l(x2) of x(2) is obtained. Gradient descent is applied to l(x1) and l(x2) on the triple loss function to solve the distance between l(x1) and l(x2), and then the text can be found by finding matching pairs. The predicted travel route is given. l(*) indicates that the convolutional network trained by cross entropy is used to extract the features of the text. Specifically, the matching pairs of features and travel routes are used to train the convolutional network. The input during training is a text x1 and a travel route x2, which are then fed to the convolutional layer, pooling layer and fully connected layer in sequence. The specific convolutional network is ResNet-100. The input during testing is a text. Therefore, the features of the text are extracted to give the predicted travel route based on the text. f(*) indicates that the Bert network is used to perform semantic segmentation on the text. concat indicates that the text features after semantic segmentation and the word segmentation data of the speech text are merged into one-dimensional features. fusion(*) indicates that the travel routes predicted by the two networks are weighted to obtain the final travel route s.

2. According to claim 1, a self-service bus management method based on bus digital transformation is characterized in that: The input method of the first interest word in S1 includes: text input method and voice recognition input method; the voice recognition input method includes: collecting user audio data, performing audio analysis on the audio data to obtain corresponding voice text information and voice text word segmentation data; the calculation formula of audio analysis is: fa=f(l n (f o (o),f(p))) (1) Among them, o represents user audio data, fa represents the word segmentation data of speech text, p represents the standard Chinese pronunciation library, l n (*) indicates the features extracted by the twin network trained with contrast loss. The matching pairs of pronunciation and text are used to train the twin network. The features extracted by the twin network give the corresponding text according to the audio. f(*) indicates the semantic segmentation of text using the Bert network. f o (*) indicates that the audio is denoised; Among them, the speech recognition is to obtain the recognition result after matching the user's pronunciation and text; that is, the twin network is used to extract audio with similar pronunciation to the text from the user's speech after denoising the speech, and the content in the audio is recognized through the text table entered in the Bert network, and finally the recognition result is obtained; the twin network includes two sub-networks, the input of the first sub-network is a pronunciation x1 in the user's audio data, which is then fed to the convolution layer, the pooling layer and the fully connected layer in sequence, and finally a feature vector f(x1) is output; the final vector l(x1) is the encoding of the input x1, the pronunciation x2 in the standard Chinese pronunciation library is input to the second sub-network, it is processed in the same way, and the encoding l(x2) of x(2) is obtained, and the distance between l(x1) and l(x2) is solved by applying gradient descent on the triple loss function, and then the text corresponding to the pronunciation in the standard Chinese pronunciation library corresponding to the user audio can be obtained by finding the matching pair.

3. According to claim 1, a self-service bus management method based on bus digital transformation is characterized in that: The S2 performs semantic association based on the first interest word, including: pre-matching a plurality of semantic association lists, wherein the semantic association lists are based on the travel destination as a keyword, and the types of public transportation services that can be provided by the destination are associated; the types of public transportation services include: transfer reminder services, travel services, catering services, accommodation services and shopping services.

4. The self-service bus management method based on bus digital transformation according to claim 1 is characterized in that: The performing semantic association based on the first interest words in S2 further includes: performing semantic association on the interest words based on a fuzzy matching algorithm, and if the interest words have words that cannot be directly matched, obtaining the words with the highest matching degree.

5. The self-service bus management method based on bus digital transformation according to claim 1 is characterized in that: The navigation of the user in S4 includes: obtaining in real time the distance between the user's current location and the nearest bus stop on the route to be taken, calculating the spatial distance between the passenger and the stop, and calculating the passenger's arrival time and the arrival time of the bus on the route to be taken.

Images