WO2020151309A1 - 最短路线计算方法、装置、计算机设备和存储介质 - Google Patents

最短路线计算方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2020151309A1
WO2020151309A1 PCT/CN2019/116647 CN2019116647W WO2020151309A1 WO 2020151309 A1 WO2020151309 A1 WO 2020151309A1 CN 2019116647 W CN2019116647 W CN 2019116647W WO 2020151309 A1 WO2020151309 A1 WO 2020151309A1
Authority
WO
WIPO (PCT)
Prior art keywords
hidden state
city
coordinates
encoding
neural network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/116647
Other languages
English (en)
French (fr)
Inventor
陈娴娴
来蕾
阮晓雯
徐亮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2020151309A1 publication Critical patent/WO2020151309A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/40—Business processes related to the transportation industry

Definitions

  • This application relates to a shortest route calculation method, device, computer equipment and storage medium.
  • a shortest route calculation method, device, computer equipment, and storage medium that can improve the accuracy of the shortest route calculation between multiple cities.
  • a calculation method for the shortest route including:
  • a sequence of shortest routes between multiple cities is generated according to the route probability.
  • a shortest route calculation device including:
  • a coordinate acquisition module configured to acquire corresponding city coordinates according to the city identifier
  • the state calculation module is configured to call a neural network model according to the route calculation task, and use the neural network model to calculate the encoding hidden state and the decoding hidden state corresponding to the city coordinates;
  • the probability calculation module is configured to use the neural network model to calculate the route probability of each city identifier according to the encoding hidden state, decoding hidden state, and preset relationship corresponding to the city coordinates;
  • the sequence generation module is used to generate the shortest route sequence between multiple cities according to the route probability.
  • a computer device including a memory and one or more processors, the memory stores computer readable instructions, and when the computer readable instructions are executed by the processor, the one or more processors execute The following steps:
  • a sequence of shortest routes between multiple cities is generated according to the route probability.
  • One or more non-volatile computer-readable storage media storing computer-readable instructions.
  • the computer-readable instructions When executed by one or more processors, the one or more processors perform the following steps:
  • a sequence of shortest routes between multiple cities is generated according to the route probability.
  • Fig. 1 is an application scenario diagram of the shortest route calculation method in one or more embodiments.
  • Fig. 2 is a schematic flow chart of the shortest route calculation method in one or more embodiments.
  • Fig. 3 is a schematic flowchart of the steps of calculating the route probability of each city identifier in one or more embodiments.
  • Fig. 4 is a block diagram of the shortest route calculation device in one or more embodiments.
  • Figure 5 is a block diagram of a computer device in one or more embodiments.
  • the shortest route calculation method provided in this application can be applied to the application scenario shown in FIG. 1.
  • the terminal 102 communicates with the server 104 through the network.
  • the server 104 obtains the route calculation task uploaded by the terminal 102, and the route calculation task carries multiple city identifiers.
  • the server 104 obtains the corresponding city coordinates according to the city identifier.
  • the server 104 calls the neural network model according to the route calculation task, and uses the neural network model to calculate the encoding hidden state and the decoding hidden state corresponding to the city coordinates.
  • the server 104 uses the neural network model to calculate the route probability of each city identifier according to the encoding hidden state, the decoding hidden state and the preset relationship corresponding to the city coordinates.
  • the server 104 generates the shortest route sequence between multiple cities according to the route probability, so that the length of the output shortest route sequence corresponds to the length of the city sequence generated by the city coordinates in a one-to-one correspondence. This can effectively solve the problem of the output sequence lengthening caused by the heuristic algorithm, and improve the accuracy of the shortest route calculation between multiple cities.
  • the terminal 102 may be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices.
  • the server 104 may be implemented as an independent server or a server cluster composed of multiple servers.
  • a method for calculating the shortest line is provided. Taking the method applied to the server in FIG. 1 as an example, the method includes the following steps:
  • Step 202 Obtain a route calculation task.
  • the route calculation task carries multiple city identifiers.
  • Step 204 Obtain corresponding city coordinates according to the city identifier.
  • the server obtains the route calculation task, and the route calculation task may be uploaded by the terminal or created by the server.
  • the server analyzes the route calculation task to obtain multiple city identifiers carried by the route calculation task.
  • the city identifier can be any one or a combination of characters such as numbers, characters, letters, etc.
  • the server obtains the corresponding city coordinates according to the city identifier, and the city coordinates may be geographic coordinates.
  • the server generates a city sequence based on the city coordinates in the order of the city identifiers.
  • the order of the city coordinates in the city sequence represents the order in which the server uses the neural network model to encode the city coordinates. .
  • the server obtains the route calculation task uploaded by the terminal, analyzes the route calculation task, and obtains five digital identifiers 1, 2, 3, 4, and 5 carried by the route calculation task.
  • the origin is the city represented by the digital identifier 1.
  • the server obtains the corresponding city coordinates according to the digital ID, and generates the city sequence [[0,0],[0,1],[-1,1], the city coordinates in the order of the city IDs 1, 2, 3, 4, and 5 [1,1], [0,2]].
  • the server uses the neural network model to sequentially encode the city coordinates in the order of the city coordinates in the city sequence.
  • step 206 the neural network model is invoked according to the route calculation task, and the neural network model is used to calculate the encoding hidden state and the decoding hidden state corresponding to the city coordinates.
  • the neural network model can be called according to the route calculation task, and the neural network model is used to calculate the encoding hidden state and decoding hidden state corresponding to the city coordinates. .
  • the server calls the neural network model according to the route calculation task.
  • the neural network model includes an initial hidden state, and the initial hidden state includes an initial encoding hidden state and an initial decoding hidden state.
  • the initial encoding hidden state refers to an initial state information when the server encodes the city coordinates through the neural network model.
  • the initial decoding hidden state refers to the initial state information when the server decodes the encoded hidden states corresponding to multiple city coordinates after the city coordinates are encoded through the neural network model.
  • the server uses the initial encoding hidden state to encode the city coordinates through the neural network model, and obtains the encoding hidden state corresponding to the city coordinates.
  • the encoding hidden state refers to the state information output by the server in the process of encoding the city coordinates through the neural network model.
  • the server encodes all the city coordinates through the neural network model, the encoding hidden state corresponding to the multiple city coordinates is transmitted to the decoding process as a semantic vector, and the encoding process ends at this time.
  • the decoding process includes multiple decoding links, and the decoding links correspond to the city identifiers.
  • the server uses the neural network model to use the initial decoding hidden state to decode the encoded hidden states corresponding to multiple city coordinates to obtain the decoded hidden states corresponding to the city coordinates.
  • the encoded hidden states corresponding to multiple city coordinates can be used as semantic vectors, which are referred to as semantic vectors in the following.
  • the decoding hidden state refers to the state information output by the server in the process of decoding the semantic vector through the neural network model.
  • step 208 the neural network model is used to calculate the route probability of each city identifier according to the encoding hidden state, the decoding hidden state and the preset relationship corresponding to the city coordinates.
  • the server uses the neural network model to calculate the encoding hidden state and the decoding hidden state corresponding to the city coordinates, it can use the neural network model to calculate the route probability of each city identifier according to the encoding hidden state, the decoding hidden state and the preset relationship corresponding to the city coordinates.
  • the route probability refers to the probability of outputting each city identifier in each decoding link in the process of the server using the neural network model to decode the semantic vector. It can also be called the attention distribution probability.
  • the preset relationship may be the relationship between the encoding hidden state corresponding to the city coordinates and the decoding hidden state corresponding to the city coordinates, and the preset relationship may be represented by a similarity function.
  • the server uses the neural network model to decode the semantic vector, and each time a decoded hidden state is output, the decoded hidden state is compared with the encoded hidden state corresponding to multiple city coordinates in the encoding process to obtain the corresponding information for each city coordinate.
  • the server normalizes the similarity to obtain the route probability of each city identifier.
  • the normalization processing method can use the softmax function.
  • Step 210 Generate a shortest route sequence between multiple cities according to the route probability.
  • the server After the server calculates the route probability of each city identifier, it can generate the shortest route sequence between multiple cities according to the route probability. Specifically, the server compares the decoding hidden state corresponding to each city coordinate with the encoding hidden state corresponding to multiple city coordinates in the encoding process, and calculates the decoding hidden state corresponding to each city coordinate and the encoding hidden state corresponding to multiple city identifiers. The similarity of the state. The server generates the attention pointer corresponding to each city identifier by using the decoded hidden state and the corresponding similarity, and outputs the city identifier corresponding to the maximum probability according to the maximum probability of the attention pointer pointing to the route probability. In the decoding process, each decoding link will output a city identifier corresponding to the maximum probability, and generate the shortest route sequence between multiple cities according to the multiple city identifiers corresponding to the maximum probability.
  • the server obtains the corresponding city coordinates according to the digital ID, and generates the city sequence [[0,0],[0,1],[-1,1],[1,1] from multiple city coordinates in the order of the city ID. , [0,2]].
  • the server calls the neural network model according to the route calculation task, and encodes the city coordinates in the order of the city coordinates in the city sequence through the neural network model using the initial encoding hidden state.
  • the server uses the initial coding hidden state e 0 to encode the city coordinates [0, 0] of the No. 1 city through the neural network model, and outputs the coding hidden state e 1 .
  • the server uses the coding hidden state e 1 to encode the city coordinates [0,1] of the 2nd city through the neural network model, and outputs the coding hidden state e 2 .
  • the city coordinates are coded according to the above coding process.
  • the server uses the coding hidden state e 4 to code the city coordinates [0, 2] of the 5th city through the neural network model, and outputs the coded hidden state s 5 .
  • the server may transmit the encoding hidden state e 5 as the semantic vector c to the encoding process, at which time the encoding process ends.
  • the server After the encoding process is over, proceed to the decoding process.
  • the server reads the initial decoding hidden state d 0 and the city identifier 1 of the originating station through the neural network model, decodes the semantic vector c, and outputs the decoded hidden state d 1 .
  • the decoding hidden state d 1 is compared with the encoding hidden state ⁇ e 1 , e 2 , e 3 , e 4 , e 5 ⁇ in the encoding process, and the corresponding similarity is calculated according to the similarity function.
  • the similarity is normalized to obtain the route probability ⁇ 0.1,0.2,0.2,0.4,0.1 ⁇ of each city identification. Select the No. 4 city corresponding to the maximum probability of 0.4 for output.
  • the server uses the neural network model to read the city identifier 4 and decode the hidden state d1 to decode the semantic vector c, and output the decoded hidden state d2.
  • the decoding hidden state d2 is compared with the encoding hidden state ⁇ e 1 , e 2 , e 3 , e 4 , e 5 ⁇ in the encoding process, and the corresponding similarity is calculated according to the similarity calculation formula.
  • the similarity is normalized to obtain the route probability ⁇ 0.05,0.4,0.1,0.3,0.15 ⁇ of each city identification, and the second city corresponding to the maximum probability of 0.4 is selected for output.
  • the semantic vector is decoded to obtain the shortest route sequence ⁇ 1,4,2,5,3 ⁇ between multiple cities.
  • the neural network model is used to calculate the encoding hidden state and the decoding hidden state corresponding to the city coordinates, and the neural network model is used to calculate the route of each city identification according to the encoding hidden state, decoding hidden state and the preset relationship corresponding to the city coordinates. Probability.
  • the shortest route sequence between multiple cities is generated according to the route probability of each city identifier, so that the length of the output shortest route sequence corresponds to the length of the acquired city sequence one-to-one. This can effectively solve the problem of the output sequence lengthening caused by the heuristic algorithm, and improve the accuracy of the shortest route calculation between multiple cities.
  • the neural network model includes the initial encoding hidden state and the initial decoding hidden state.
  • Using the neural network model to calculate the encoding hidden state and the decoding hidden state corresponding to the city coordinates includes: Coordinates are encoded to obtain the encoded hidden state corresponding to the city coordinates; the neural network model uses the initial decoding hidden state to decode the encoded hidden state corresponding to the city coordinates to obtain the decoded hidden state corresponding to the city coordinates.
  • the server sequentially encodes the city coordinates through the neural network model.
  • the encoding process includes multiple encoding links, and the encoding links correspond to the city coordinates one by one.
  • the server uses the neural network model to use the initial decoding hidden state to encode any city coordinate, output the encoding hidden state corresponding to the city coordinate, and transmit the encoded hidden state to the next encoding link.
  • the city coordinates are coded to obtain multiple coded hidden states.
  • the server reads the initial decoded hidden state and the city identifier of the originating station through the neural network model, decodes the encoded hidden state corresponding to the city coordinates, outputs the decoded hidden state corresponding to the city coordinates, and transmits the decoded hidden state to the next decoding link.
  • the server cyclically decodes the semantic vector through the neural network model to obtain the decoded hidden state corresponding to the city coordinate.
  • the server uses the neural network model to use the initial encoding hidden state to encode the city coordinates and the initial decoding hidden state to decode the encoding hidden state corresponding to the city coordinates, which can more accurately analyze the city coordinates and effectively Improve the accuracy of the shortest route calculation.
  • the neural network model encodes the city coordinates through multiple encoding links, and the initial encoding hidden state is used to encode the city coordinates through the neural network model, and obtaining the encoding hidden state corresponding to the city coordinates includes: through the neural network The model selects the target coordinates in the city coordinates; the neural network model uses the initial decoding hidden state to encode the target coordinates, outputs the encoded hidden state corresponding to the target coordinates, and transmits the encoded hidden state to the next encoding link; according to the neural network model The encoding hidden state encodes the next city coordinate in the city coordinates to obtain the next encoding hidden state; repeat the steps of encoding the city coordinates with the initial encoding hidden state through the neural network model until the encoding hidden state corresponding to the city coordinates is obtained .
  • the server can select any city coordinate in the city coordinates as the target coordinate through the neural network model, encode the target coordinate, output the encoded hidden state corresponding to the target coordinate, and transmit the encoded hidden state to the next encoding link through the neural network
  • the model encodes the next city coordinate in the city coordinates according to the encoding hidden state to obtain the next encoding hidden state.
  • the server repeats the above steps of encoding city coordinates using the initial encoding hidden state through the neural network model, until all the encoding hidden states corresponding to the city coordinates are obtained. By calculating one by one the coding hidden states corresponding to the city coordinates, the accuracy of subsequent route calculations can be further improved.
  • the neural network model decodes the city coordinates through multiple decoding links.
  • the neural network model uses the initial decoding hidden state to decode the encoded hidden state corresponding to the city coordinates to obtain the decoded hidden state corresponding to the city coordinates.
  • the server reads the initial decoding hidden state and the encoding hidden state corresponding to the city coordinates of the originating station through the neural network model for decoding, outputs the decoding hidden state corresponding to the city coordinates, and transmits the decoded hidden state to the next decoding link.
  • the server decodes the next encoding hidden state according to the decoding hidden state through the neural network model, and repeats the above steps of using the neural network model to decode the encoding hidden state corresponding to the city coordinates using the initial decoding hidden state, until the decoding corresponding to the city coordinates is obtained Hidden state. It can decode according to the coding order of the city coordinates, so that the decoding hidden state corresponding to the city coordinates corresponds to the number of city coordinates and the coding link one to one, thereby effectively solving the problem of the output sequence becoming longer.
  • the above method further includes the step of using a neural network model to calculate the route probability of each city identifier according to the encoding hidden state, decoding hidden state and preset relationship corresponding to the city coordinates, as shown in FIG.
  • the specific steps include:
  • Step 302 Compare the decoded hidden state corresponding to each city coordinate with the encoded hidden state corresponding to multiple city coordinates to obtain the similarity between the decoded hidden state corresponding to each city coordinate and the encoded hidden state corresponding to multiple city identifiers .
  • Step 304 Normalize the similarity to obtain the route probability of each city identifier.
  • the server may compare the decoding hidden state corresponding to each city coordinate with the encoding hidden state corresponding to multiple city coordinates. Specifically, in the process of decoding using the neural network model, the server compares the decoding hidden state with the encoding hidden state corresponding to multiple city coordinates in the encoding process every time it outputs a decoding hidden state, and calculates the decoding hidden state and The similarity of each coded hidden state. Among them, the similarity can be calculated by the following formula:
  • i is the order of the current decoded hidden state
  • j is the number of encoded hidden states
  • v is model parameters
  • T is a vector transposition function
  • e j is the j-th encode hidden state
  • d i is the i-th decoding hidden state
  • W 1 is a model parameter
  • W 2 for the model parameters.
  • the similarity can be normalized to obtain the route probability of each city identifier.
  • the normalization processing method can use the softmax function.
  • the route probability can be calculated using the following formula:
  • C 1 ,..., C i-1 , P is the route probability of each city identification.
  • a neural network model corresponding to multiple cities can be established, such as a Sequence to Sequence model (sequence conversion model, Seq2Seq model for short) to solve the problem of the output sequence becoming longer.
  • a Sequence to Sequence model sequence conversion model, Seq2Seq model for short
  • the length of the output sequence changes accordingly, which makes the neural network model no longer applicable. Therefore, in the face of changes in the number of cities, there is a problem with the generality of the neural network model.
  • the length of the output sequence corresponds to the length of the city sequence generated by the city coordinates. This solves the problem that the length of the output sequence that ordinary neural networks cannot solve depends on the length of the input sequence. Train the training set after the number of cities is changed to solve the shortest route calculation problem for different numbers of cities.
  • generating the shortest route sequence between multiple cities according to the route probability includes: using the decoded hidden state and the corresponding similarity to generate the attention pointer corresponding to each city identifier; according to the attention pointer pointing to the route probability The maximum probability in; output the city identifier corresponding to the maximum probability.
  • the server compares the decoded hidden state corresponding to each city coordinate with the encoded hidden state corresponding to multiple city coordinates. During the comparison process, an attention pointer corresponding to each city identifier is generated. Attention refers to The ability to select a small part of useful information from a large amount of input information to focus on processing and ignore other information.
  • the similarity between the decoded hidden state corresponding to the city coordinates and the encoded hidden state corresponding to multiple city identifiers is calculated, and the similarity is normalized to obtain the route probability corresponding to each city identifier.
  • the server selects the maximum probability among the route probabilities according to the attention pointer, and outputs the city identifier corresponding to the maximum probability, thereby obtaining the shortest route sequence generated by multiple city identifiers corresponding to the maximum probability.
  • the server points to the maximum probability of the route probability according to the attention pointer, and does not need to consider the city coordinates in the city sequence.
  • the shortest route sequence can be obtained.
  • the length of the shortest route sequence corresponds to the length of the city sequence generated by the city coordinates, which solves the serious dependence of the output sequence length that cannot be solved by ordinary neural networks. For problems such as the length of the input sequence, it is not necessary to train the training set after the number of cities is changed, so as to solve the shortest route calculation problem for different numbers of cities.
  • the above method further includes: storing the location information of the encoding hidden state corresponding to the city coordinates; performing the decoding hidden state corresponding to each city coordinate with the encoding hidden state corresponding to multiple city coordinates according to the location information. Compare, get the similarity corresponding to the hidden state of decoding.
  • the server uses the neural network model to encode the city coordinates
  • the location information of the encoding hidden state is identified, and the location information is stored.
  • the server transmits the location information of each coded hidden state to the decoding process.
  • the decoding hidden state corresponding to each city coordinate is coded with multiple city coordinates according to the location information of the coded hidden state.
  • the hidden state is compared to obtain the similarity corresponding to the decoded hidden state.
  • the server transmits the position information of each encoded hidden state to the decoding process, which shortens the information transmission distance and effectively improves the comparison efficiency.
  • the above method further includes: obtaining the preset shortest route sequence according to city coordinates; and calculating the loss of the shortest route sequence according to the city coordinates, the preset shortest route sequence, the shortest route sequence, and the preset relationship.
  • the server obtains the preset shortest route sequence according to the city coordinates, generates the city sequence according to the order of the city identifiers from the city coordinates, and calculates the loss of the shortest route sequence using the city coordinate sequence and the preset relationship.
  • the preset relationship may be the relationship between the preset shortest route sequence and the shortest route sequence, and the preset relationship may be represented by a loss function.
  • the loss of the shortest route sequence can be calculated by the following formula:
  • C is the loss of the shortest route sequence
  • n is the total number of city coordinates
  • x is the acquired city coordinate
  • y is the city coordinate in the preset shortest route sequence
  • a is the city coordinate in the shortest route sequence.
  • the server calculates the loss of the shortest route sequence by obtaining the preset shortest route sequence. On this basis, the server can train the parameters through the neural network model to reduce the loss of the shortest route sequence and effectively improve the shortest route sequence. Calculation accuracy.
  • a device for calculating the shortest line including: a communication module 402, a coordinate acquisition module 404, a state calculation module 406, a probability calculation module 408, and a sequence generation module 410, wherein:
  • the communication module 402 is used to obtain a route calculation task, and the route calculation task carries multiple city identifiers.
  • the coordinate acquisition module 404 is configured to acquire corresponding city coordinates according to the city identifier.
  • the state calculation module 406 is configured to call the neural network model according to the route calculation task, and use the neural network model to calculate the encoding hidden state and the decoding hidden state corresponding to the city coordinates.
  • the probability calculation module 408 is configured to use the neural network model to calculate the route probability of each city identifier according to the encoding hidden state, the decoding hidden state and the preset relationship corresponding to the city coordinates.
  • the sequence generating module 410 is used to generate the shortest route sequence between multiple cities according to the route probability.
  • the state calculation module 406 is used to encode the city coordinates using the initial encoding hidden state through the neural network model to obtain the encoding hidden state corresponding to the city coordinates; the neural network model uses the initial decoding hidden state to encode the city coordinates The corresponding encoded hidden state is decoded to obtain the decoded hidden state corresponding to the city coordinates.
  • the state calculation module 406 is also used to select the target coordinates in the city coordinates through the neural network model; the neural network model uses the initial decoding hidden state to encode the target coordinates, and output the encoded hidden state corresponding to the target coordinates , The coding hidden state is transmitted to the next coding link; the next city coordinate in the city coordinates is coded according to the coding hidden state through the neural network model to obtain the next coding hidden state; the neural network model is repeatedly used to pair the initial coding hidden state The steps of encoding the city coordinates until the encoding hidden state corresponding to the city coordinates is obtained.
  • the state calculation module 406 is also used to decode the encoded hidden state corresponding to the target coordinate by using the initial decoding hidden state through the neural network model to obtain the decoded hidden state corresponding to the target coordinate, and transmit the decoded hidden state to the next A decoding link; decode the next coded hidden state according to the decoding hidden state through the neural network model; repeat the steps of decoding the coded hidden state corresponding to the city coordinates through the neural network model using the initial decoding hidden state, until the corresponding city coordinates are obtained The hidden state of decoding.
  • the probability calculation module 408 is used to compare the decoded hidden state corresponding to each city coordinate with the encoded hidden state corresponding to multiple city coordinates to obtain the decoded hidden state corresponding to each city coordinate and multiple The similarity of the coding hidden state corresponding to the city identifier; normalize the similarity to obtain the route probability of each city identifier.
  • the sequence generation module 410 is used to generate the attention pointer corresponding to each city identifier by using the decoding hidden state and the corresponding similarity; according to the maximum probability of the attention pointer pointing to the route probability, the maximum probability corresponds to The city logo is output.
  • the above-mentioned device further includes a similarity calculation module, the similarity calculation module is used to store the location information of the encoding hidden state corresponding to the city coordinates; according to the location information, the decoding hidden state corresponding to each city coordinate Comparing the coding hidden states corresponding to multiple city coordinates to obtain the similarity corresponding to the decoding hidden states.
  • the above device further includes a loss calculation module for obtaining a preset shortest route sequence according to city coordinates; calculating the shortest route according to city coordinates, preset shortest route sequence, shortest route sequence, and preset relationship Loss of route sequence.
  • Each module in the above-mentioned shortest route calculation device can be implemented in whole or in part by software, hardware, and a combination thereof.
  • the foregoing modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the foregoing modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 5.
  • the computer equipment includes a processor, a memory, a network interface and a database connected by a system bus. Among them, the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system, computer readable instructions, and a database.
  • the internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium.
  • the database of the computer equipment is used to store the state information corresponding to the city coordinates.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer readable instruction is executed by the processor to realize a shortest route calculation method.
  • FIG. 5 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or less parts than shown in the figure, or combining some parts, or having a different part arrangement.
  • a computer device including a memory and one or more processors.
  • the memory stores computer-readable instructions.
  • the one or more processors execute the above-mentioned methods. The steps in the example.
  • One or more non-volatile computer-readable storage media storing computer-readable instructions.
  • the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps in each of the foregoing method embodiments. step.
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • ROM read only memory
  • PROM programmable ROM
  • EPROM electrically programmable ROM
  • EEPROM electrically erasable programmable ROM
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Tourism & Hospitality (AREA)
  • General Physics & Mathematics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Navigation (AREA)

Abstract

一种最短路线计算方法，包括：获取路线计算任务，路线计算任务携带多个城市标识；根据城市标识获取对应的城市坐标；根据路线计算任务调用神经网络模型，利用神经网络模型计算城市坐标对应的编码隐状态以及解码隐状态；利用神经网络模型根据城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率；根据路线概率生成多个城市之间的最短路线序列。

Description

最短路线计算方法、装置、计算机设备和存储介质
本申请要求于2019年1月22日提交中国专利局，申请号为2019100563432，申请名称为“最短路线计算方法、装置、计算机设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及一种最短路线计算方法、装置、计算机设备和存储介质。
背景技术
在工业应用中，会经常遇到交通运输及物流配送等问题。为了节省时间及提高运输效率，就需要规划经过的多个城市之间的最短路线。多个城市之间的最短路线通常采用启发式算法进行求解，但是启发式算法又容易引发输出序列变长的问题，导致多个城市之间的最短路线计算不准确。因此，如何提高多个城市之间的最短路线计算的准确性成为目前需要解决的一个技术问题。
发明内容
根据本申请公开的各种实施例，提供一种能够提高多个城市之间的最短路线计算的准确性的最短路线计算方法、装置、计算机设备和存储介质。
一种最短路线计算方法，包括：
获取路线计算任务，所述路线计算任务携带多个城市标识；
根据所述城市标识获取对应的城市坐标；
根据所述路线计算任务调用神经网络模型，利用所述神经网络模型计算所述城市坐标对应的编码隐状态以及解码隐状态；
利用所述神经网络模型根据所述城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率；及
根据所述路线概率生成多个城市之间的最短路线序列。
一种最短路线计算装置，包括：
通信模块，用于获取路线计算任务，所述路线计算任务携带多个城市标识；
坐标获取模块，用于根据所述城市标识获取对应的城市坐标；
状态计算模块，用于根据所述路线计算任务调用神经网络模型，利用所述神经网络模型计算所述城市坐标对应的编码隐状态以及解码隐状态；
概率计算模块，用于利用所述神经网络模型根据所述城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率；及
序列生成模块，用于根据所述路线概率生成多个城市之间的最短路线序列。
一种计算机设备，包括存储器和一个或多个处理器，所述存储器中储存有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述一个或多个处理器执行以下步骤：
获取路线计算任务，所述路线计算任务携带多个城市标识；
根据所述城市标识获取对应的城市坐标；
根据所述路线计算任务调用神经网络模型，利用所述神经网络模型计算所述城市坐标对应的编码隐状态以及解码隐状态；
利用所述神经网络模型根据所述城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率；及
根据所述路线概率生成多个城市之间的最短路线序列。
一个或多个存储有计算机可读指令的非易失性计算机可读存储介质，计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行以下步骤：
获取路线计算任务，所述路线计算任务携带多个城市标识；
根据所述城市标识获取对应的城市坐标；
根据所述路线计算任务调用神经网络模型，利用所述神经网络模型计算所述城市坐标对应的编码隐状态以及解码隐状态；
利用所述神经网络模型根据所述城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率；及
根据所述路线概率生成多个城市之间的最短路线序列。
本申请的一个或多个实施例的细节在下面的附图和描述中提出。本申请的其它特征和优点将从说明书、附图以及权利要求书变得明显。
附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通 技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其它的附图。
图1为一个或多个实施例中最短路线计算方法的应用场景图。
图2为一个或多个实施例中最短路线计算方法的流程示意图。
图3为一个或多个实施例中计算每个城市标识的路线概率步骤的流程示意图。
图4为一个或多个实施例中最短路线计算装置的框图。
图5为一个或多个实施例中计算机设备的框图。
具体实施方式
为了使本申请的技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请提供的最短路线计算方法，可以应用于如图1所示的应用场景中。终端102通过网络与服务器104进行通信。服务器104获取终端102上传的路线计算任务，路线计算任务携带多个城市标识。服务器104根据城市标识获取对应的城市坐标。服务器104根据路线计算任务调用神经网络模型，利用神经网络模型计算城市坐标对应的编码隐状态以及解码隐状态。服务器104利用神经网络模型根据城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率。服务器104根据路线概率生成多个城市之间的最短路线序列，使输出的最短路线序列的长度与城市坐标生成的城市序列的长度一一对应。由此能够有效解决启发式算法引发的输出序列变长的问题，提高了多个城市之间的最短路线计算的准确性。终端102可以但不限于是各种个人计算机、笔记本电脑、智能手机、平板电脑和便携式可穿戴设备。服务器104可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
在其中一个实施例中，如图2所示，提供了一种最短线路计算方法，以该方法应用于图1中的服务器为例进行说明，包括以下步骤：
步骤202，获取路线计算任务，路线计算任务携带多个城市标识。
步骤204，根据城市标识获取对应的城市坐标。
服务器获取路线计算任务，路线计算任务可以是终端上传的，也可以是服务器创建的。服务器对路线计算任务进行解析，得到路线计算任务携带的多个城市标识。其中，城市标识可以是数字、文字、字母等字符中的任意一种或多种的组合。服务器根据城市标识获取对应的城市坐标，城市坐标可以是地理坐标。服务器将城市坐标按照城市标识的顺序生成城市序列，城市序列中城市坐标的顺序表示服务器利用神经网络模型对城市坐标进行编码的顺序。。例如，服务器获取终端上传的路线计算任务，对路线计算任务解析，得到路线计算任务携带 的五个数字标识1、2、3、4、5，始发站为数字标识1代表的城市。服务器根据数字标识获取对应的城市坐标，将城市坐标按照城市标识1、2、3、4、5的顺序生成城市序列[[0,0],[0,1],[-1,1],[1,1]，[0,2]]。服务器利用神经网络模型按照城市序列中城市坐标的顺序依次对城市坐标进行编码。
步骤206，根据路线计算任务调用神经网络模型，利用神经网络模型计算城市坐标对应的编码隐状态以及解码隐状态。
服务器在根据城市标识获取对应的城市坐标，将城市坐标按照城市标识的顺序生成城市序列后，可根据路线计算任务调用神经网络模型，利用神经网络模型计算城市坐标对应的编码隐状态以及解码隐状态。具体的，服务器根据路线计算任务调用神经网络模型，神经网络模型包括初始隐状态，初始隐状态包括初始编码隐状态以及初始解码隐状态。初始编码隐状态是指服务器通过神经网络模型对城市坐标进行编码时的一个初始化的状态信息。初始解码隐状态是指服务器通过神经网络模型对城市坐标编码完成后，对多个城市坐标对应的编码隐状态进行解码时的一个初始化的状态信息。
在编码过程中，服务器通过神经网络模型利用初始编码隐状态对城市坐标进行编码，得到与城市坐标对应的编码隐状态。其中，编码隐状态是指服务器通过神经网络模型对城市坐标进行编码的过程中输出的状态信息。当服务器通过神经网络模型对所有的城市坐标编码完成后，将多个城市坐标对应的编码隐状态作为语义向量传送至解码过程，此时编码过程结束。
在编码过程结束后，进入解码过程，解码过程包括多个解码环节，解码环节与城市标识一一对应。服务器通过神经网络模型利用初始解码隐状态对多个城市坐标对应的编码隐状态进行解码，得到与城市坐标对应的解码隐状态。可以将多个城市坐标对应的编码隐状态作为语义向量，以下称为语义向量。其中，解码隐状态是指服务器通过神经网络模型对语义向量进行解码的过程中输出的状态信息。
步骤208，利用神经网络模型根据城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率。
服务器在利用神经网络模型计算城市坐标对应的编码隐状态以及解码隐状态后，可利用神经网络模型根据城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率。其中，路线概率是指服务器利用神经网络模型对语义向量进行解码的过程中，每个解码环节输出每个城市标识的概率，也可以称为注意力分配概率。预设关系可以是城市坐标对应的编码隐状态与城市坐标对应的解码隐状态之间的关系，预设关系可以用相似度函数来表示。具体的，服务器利用神经网络模型对语义向量进行解码，每输出一个解码隐状态，将该解码隐状态与编码过程中多个城市坐标对应的编码隐状态进行比对，得到每个城市坐标 对应的解码隐状态与多个城市标识对应的编码隐状态的相似度。服务器将相似度进行归一化处理，得到每个城市标识的路线概率。其中，归一化处理方式可以利用softmax函数。
步骤210，根据路线概率生成多个城市之间的最短路线序列。
服务器计算每个城市标识的路线概率后，可根据路线概率生成多个城市之间的最短路线序列。具体的，服务器将每个城市坐标对应的解码隐状态与编码过程中多个城市坐标对应的编码隐状态进行比对，计算每个城市坐标对应的解码隐状态与多个城市标识对应的编码隐状态的相似度。服务器利用解码隐状态与对应的相似度生成每个城市标识对应的注意力指针，根据注意力指针指向所述路线概率中的最大概率，将最大概率对应的城市标识进行输出。在解码过程中，每个解码环节会输出一个最大概率对应的城市标识，根据多个最大概率对应的城市标识生成多个城市之间的最短路线序列。
例如，服务器根据数字标识获取对应的城市坐标，将多个城市坐标按照城市标识的顺序生成城市序列[[0,0],[0,1],[-1,1],[1,1]，[0,2]]。在编码过程中，服务器根据路线计算任务调用神经网络模型，通过神经网络模型利用初始编码隐状态按照城市序列中城市坐标的顺序对城市坐标进行编码。在第一个编码环节，服务器通过神经网络模型利用初始编码隐状态e 0对1号城市的城市坐标[0,0]进行编码，输出编码隐状态e 1。在第二个编码环节，服务器通过神经网络模型利用编码隐状态e 1对2号城市的城市坐标[0,1]进行编码，输出编码隐状态e 2。根据上述编码过程对城市坐标进行编码，在第五个编码环节，服务器通过神经网络模型利用编码隐状态e 4对5号城市的城市坐标[0,2]进行编码，输出编码隐状态s 5。服务器可以将编码隐状态e 5作为语义向量c传送至编码过程，此时编码过程结束。
在编码过程结束后，进行进入解码过程。在第一个解码环节，服务器通过神经网络模型读取初始解码隐状态d 0以及始发站的城市标识1对语义向量c进行解码，输出解码隐状态d 1。将解码隐状态d 1与编码过程中的编码隐状态{e 1,e 2,e 3,e 4,e 5}进行比对，根据相似度函数计算对应的相似度。将相似度进行归一化处理，得到每个城市标识的路线概率{0.1,0.2,0.2,0.4,0.1}。选取最大概率0.4对应的4号城市进行输出。在第二个解码环节，服务器通过神经网络模型读取城市标识4以及解码隐状态d1对语义向量c进行解码，输出解码隐状态d2。将解码隐状态d2与编码过程中的编码隐状态{e 1,e 2,e 3,e 4,e 5}进行比对，根据相似度计算公式计算对应的相似度。将相似度进行归一化处理，得到每个城市标识的路线概率{0.05,0.4,0.1,0.3,0.15}，选取最大概率0.4对应的2号城市进行输出。根据上述解码过程对语义向量进行解码，得到多个城市之间的最短路线序列{1,4,2,5,3}。
在本实施例中，通过神经网络模型计算城市坐标对应的编码隐状态以及解码隐状态，利用神经网络模型根据城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标 识的路线概率。从而根据每个城市标识的路线概率生成多个城市之间的最短路线序列，使输出的最短路线序列的长度与获取的城市序列的长度一一对应。由此能够有效解决启发式算法引发的输出序列变长的问题，提高了多个城市之间的最短路线计算的准确性。
在其中一个实施例中，神经网络模型包括初始编码隐状态以及初始解码隐状态，利用神经网络模型计算城市坐标对应的编码隐状态以及解码隐状态包括：通过神经网络模型利用初始编码隐状态对城市坐标进行编码，得到与城市坐标对应的编码隐状态；通过神经网络模型利用初始解码隐状态对城市坐标对应的编码隐状态进行解码，得到与城市坐标对应的解码隐状态。
服务器通过神经网络模型依次对城市坐标进行编码，编码过程包括多个编码环节，编码环节与城市坐标一一对应。在第一个编码环节，服务器通过神经网络模型利用初始解码隐状态对任意一个城市坐标进行编码，输出与该城市坐标对应的编码隐状态，将编码隐状态传送至下一个编码环节，对下一个城市坐标进行编码，得到多个编码隐状态。服务器通过神经网络模型读取初始解码隐状态以及始发站的城市标识对城市坐标对应的编码隐状态进行解码，输出城市坐标对应的解码隐状态，将该解码隐状态传送至下一个解码环节。服务器通过神经网络模型对语义向量进行循环解码，得到城市坐标对应的解码隐状态。
在本实施例中，服务器通过神经网络模型利用初始编码隐状态对城市坐标进行编码以及利用初始解码隐状态对城市坐标对应的编码隐状态进行解码，能够更加准确地对城市坐标进行数据分析，有效提高最短路线计算的准确性。
在其中一个实施例中，神经网络模型通过多个编码环节对城市坐标进行编码，通过神经网络模型利用初始编码隐状态对城市坐标进行编码，得到与城市坐标对应的编码隐状态包括：通过神经网络模型在城市坐标中选取目标坐标；通过神经网络模型利用初始解码隐状态对目标坐标进行编码，输出与目标坐标对应的编码隐状态，将编码隐状态传送至下一个编码环节；通过神经网络模型根据编码隐状态对城市坐标中的下一个城市坐标进行编码，得到下一个编码隐状态；重复通过神经网络模型利用初始编码隐状态对城市坐标进行编码的步骤，直至得到与城市坐标对应的编码隐状态。
服务器可以通过神经网络模型在城市坐标中选取任意一个城市坐标作为目标坐标，对目标坐标进行编码，输出与该目标坐标对应的编码隐状态，将编码隐状态传送至下一个编码环节，通过神经网络模型根据编码隐状态对城市坐标中的对下一个城市坐标进行编码，得到下一个编码隐状态。服务器重复上述通过神经网络模型利用初始编码隐状态对城市坐标进行编码的步骤，直至得到所有城市坐标对应的编码隐状态。通过对城市坐标对应的编码隐状态进行逐个计算，能够进一步提高后续路线计算的准确性。
在其中一个实施例中，神经网络模型通过多个解码环节对城市坐标进行解码，通过神经网络模型利用初始解码隐状态对城市坐标对应的编码隐状态进行解码，得到与城市坐标对应的解码隐状态包括：通过神经网络模型利用初始解码隐状态对目标坐标对应的编码隐状态进行解码，得到目标坐标对应的解码隐状态，将解码隐状态传送至下一个解码环节；通过神经网络模型根据解码隐状态对下一个编码隐状态进行解码；重复通过神经网络模型利用初始解码隐状态对城市坐标对应的编码隐状态进行解码的步骤，直至得到与城市坐标对应的解码隐状态。
服务器通过神经网络模型读取初始解码隐状态以及始发站的城市坐标对应的编码隐状态进行解码，输出城市坐标对应的解码隐状态，将该解码隐状态传送至下一个解码环节。服务器通过神经网络模型根据解码隐状态对下一个编码隐状态进行解码，重复上述通过神经网络模型利用初始解码隐状态对城市坐标对应的编码隐状态进行解码的步骤，直至得到与城市坐标对应的解码隐状态。能够按照城市坐标的编码顺序进行解码，使城市坐标对应的解码隐状态与城市坐标的数量以及编码环节一一对应，从而有效解决输出序列变长的问题。
在其中一个实施例中，上述方法还包括：利用神经网络模型根据城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率的步骤，如图3所示，该步骤具体包括：
步骤302，将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进行比对，得到每个城市坐标对应的解码隐状态与多个城市标识对应的编码隐状态的相似度。
步骤304，将相似度进行归一化处理，得到每个城市标识的路线概率。
服务器在得到与城市坐标对应的编码隐状态以及解码隐状态后，可将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进行比对。具体的，服务器在利用神经网络模型进行解码的过程中，每输出一个解码隐状态，就将解码隐状态与编码过程中的多个城市坐标对应的编码隐状态进行比对，计算解码隐状态与每个编码隐状态的相似度。其中，相似度可以通过如下公式进行计算：
Figure PCTCN2019116647-appb-000001
其中，
Figure PCTCN2019116647-appb-000002
为第i个解码隐状态与j个城市坐标对应的编码隐状态的相似度，可以称为解码过程的激活函数，i为当前解码隐状态的顺序，j为编码隐状态的个数，v为模型参数，T为向量转置函数，e j为j个编码隐状态，d i为第i个解码隐状态，W 1为模型参数，W 2为模型参数。
服务器在计算每个城市坐标对应的解码隐状态与多个城市标识对应的编码隐状态的相似 度后，可将相似度进行归一化处理，得到每个城市标识的路线概率。其中，归一化处理方式可以利用softmax函数。路线概率可以利用如下公式进行计算：
P(C i|C 1，…，C i-1，P)＝soft max(u i)
其中，C 1，…，C i-1，P为每个城市标识的路线概率。
在传统方式中，可以建立多个城市对应的神经网络模型，例如Sequence to Sequence模型(序列转换模型，简称Seq2Seq模型)，以解决输出序列变长的问题。但是当城市数量改变时，输出序列的长度也相应改变，导致该神经网络模型不再适用。因此，在面对城市数量发生变化的情况下，神经网络模型的通用性存在问题。而在本实施例中，当城市数量改变时，只需要计算对城市坐标对应的解码隐状态与多个城市标识对应的编码隐状态的相似度，将相似度进行归一化处理，再将最大概率对应的城市标识进行输出，此时输出序列的长度与城市坐标生成的城市序列的长度一一对应，解决了普通神经网络无法解决的输出序列长度严重依赖于输入序列长度这类问题，不需要对城市数量改变后的训练集进行训练，实现解决不同城市数量的最短路线计算问题。
在其中一个实施例中，根据路线概率生成多个城市之间的最短路线序列，包括：利用解码隐状态与对应的相似度生成每个城市标识对应的注意力指针；根据注意力指针指向路线概率中的最大概率；将最大概率对应的城市标识进行输出。
服务器在将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进行比对，在进行比对的过程中会生成每个城市标识对应的注意力指针，注意力是指从大量的输入信息中选择小部分的有用信息来重点处理，并忽略其他信息的能力。在每一个解码环节，都会对城市坐标对应的解码隐状态与多个城市标识对应的编码隐状态的相似度进行计算，将相似度进行归一化处理，得到每个城市标识对应的路线概率。服务器根据注意力指针选取路线概率中的最大概率，输出最大概率对应的城市标识，从而得到多个最大概率对应的城市标识生成的最短路线序列。
在本实施例中，服务器根据注意力指针指向路线概率中的最大概率，不需要考率城市序列中的城市坐标。通过将最大概率对应的城市标识进行输出，由此可以得到最短路线序列，最短路线序列的长度与城市坐标生成的城市序列的长度一一对应，解决了普通神经网络无法解决的输出序列长度严重依赖于输入序列长度这类问题，不需要对城市数量改变后的训练集进行训练，实现解决不同城市数量的最短路线计算问题。
在其中一个实施例中，上述方法还包括：将城市坐标对应的编码隐状态的位置信息进行存储；根据位置信息将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进 行比对，得到解码隐状态对应的相似度。
服务器利用神经网络模型对城市坐标进行编码的过程中，每输出一个城市坐标对应的编码隐状态，识别将该编码隐状态的位置信息，将位置信息进行存储。服务器将每个编码隐状态的位置信息传送至解码过程，在解码过程中的每个解码环节，根据编码隐状态的位置信息将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进行比对，从而得到解码隐状态对应的相似度。
在本实施例中，服务器通过将每个编码隐状态的位置信息传送至解码过程，缩短了信息传递距离，有效提高了比对效率。
在其中一个实施例中，上述方法还包括：根据城市坐标获取预设最短路线序列；根据城市坐标、预设最短路线序列、最短路线序列以及预设关系计算最短路线序列的损失。
在本实施例中，服务器根据城市坐标获取预设最短路线序列，将城市坐标按照城市标识的顺序生成城市序列，利用城市坐标序列以及预设关系计算最短路线序列的损失。其中，预设关系可以是预设最短路线序列与最短路线序列之间的关系，预设关系可以用损失函数来表示。最短路线序列的损失可以通过如下公式进行计算：
Figure PCTCN2019116647-appb-000003
其中，C为最短路线序列的损失，n为城市坐标的总数，x为获取的城市坐标，y为预设最短路线序列中的城市坐标，a为最短路线序列中的的城市坐标。
在本实施例中，服务器通过获取预设最短路线序列，计算最短路线序列的损失，在此基础上，服务器可以通过神经网络模型来训练参数使最短路线序列的损失减小，有效提高最短路线序列计算的准确性。
应该理解的是，虽然图2-3的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，图2-3中的至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些子步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
在其中一个实施例中，如图4所示，提供了一种最短线路计算装置，包括：通信模块402、坐标获取模块404、状态计算模块406、概率计算模块408以及序列生成模块410，其中：
通信模块402，用于获取路线计算任务，路线计算任务携带多个城市标识。
坐标获取模块404，用于根据城市标识获取对应的城市坐标。
状态计算模块406，用于根据路线计算任务调用神经网络模型，利用神经网络模型计算城市坐标对应的编码隐状态以及解码隐状态。
概率计算模块408，用于利用神经网络模型根据城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率。
序列生成模块410，用于根据路线概率生成多个城市之间的最短路线序列。
在其中一个实施例中，状态计算模块406用于通过神经网络模型利用初始编码隐状态对城市坐标进行编码，得到与城市坐标对应的编码隐状态；通过神经网络模型利用初始解码隐状态对城市坐标对应的编码隐状态进行解码，得到与城市坐标对应的解码隐状态。
在其中一个实施例中，状态计算模块406还用于通过神经网络模型在城市坐标中选取目标坐标；通过神经网络模型利用初始解码隐状态对目标坐标进行编码，输出与目标坐标对应的编码隐状态，将编码隐状态传送至下一个编码环节；通过神经网络模型根据编码隐状态对城市坐标中的下一个城市坐标进行编码，得到下一个编码隐状态；重复通过神经网络模型利用初始编码隐状态对城市坐标进行编码的步骤，直至得到与城市坐标对应的编码隐状态。
在其中一个实施例中，状态计算模块406还用于通过神经网络模型利用初始解码隐状态对目标坐标对应的编码隐状态进行解码，得到目标坐标对应的解码隐状态，将解码隐状态传送至下一个解码环节；通过神经网络模型根据解码隐状态对下一个编码隐状态进行解码；重复通过神经网络模型利用初始解码隐状态对城市坐标对应的编码隐状态进行解码的步骤，直至得到与城市坐标对应的解码隐状态。
在其中一个实施例中，概率计算模块408用于将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进行比对，得到每个城市坐标对应的解码隐状态与多个城市标识对应的编码隐状态的相似度；将相似度进行归一化处理，得到每个城市标识的路线概率。
在其中一个实施例中，序列生成模块410用于利用解码隐状态与对应的相似度生成每个城市标识对应的注意力指针；根据注意力指针指向路线概率中的最大概率，将最大概率对应的城市标识进行输出。
在其中一个实施例中，上述装置还包括相似度计算模块，该相似度计算模块用于将城市坐标对应的编码隐状态的位置信息进行存储；根据位置信息将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进行比对，得到解码隐状态对应的相似度。
在其中一个实施例中，上述装置还包括损失计算模块，该损失计算模块用于根据城市坐标获取预设最短路线序列；根据城市坐标、预设最短路线序列、最短路线序列以及预设关系计算最短路线序列的损失。
关于最短路线计算装置的具体限定可以参见上文中对于最短路线计算方法的限定，在此 不再赘述。上述最短路线计算装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在其中一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图5所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为非易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储城市坐标对应的状态信息。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种最短路线计算方法。
本领域技术人员可以理解，图5中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
一种计算机设备，包括存储器及一个或多个处理器，存储器中储存有计算机可读指令，计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行上述各个方法实施例中的步骤。
一个或多个存储有计算机可读指令的非易失性计算机可读存储介质，计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行上述各个方法实施例中的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储于一非易失性计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各 个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (20)

  1. 一种最短路线计算方法，包括：
    获取路线计算任务，所述路线计算任务携带多个城市标识；
    根据所述城市标识获取对应的城市坐标；
    根据所述路线计算任务调用神经网络模型，利用所述神经网络模型计算所述城市坐标对应的编码隐状态以及解码隐状态；
    利用所述神经网络模型根据所述城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率；及
    根据所述路线概率生成多个城市之间的最短路线序列。
  2. 根据权利要求1所述的方法，其特征在于，所述神经网络模型包括初始编码隐状态以及初始解码隐状态，所述利用所述神经网络模型计算所述城市坐标对应的编码隐状态以及解码隐状态，包括：
    通过所述神经网络模型利用所述初始编码隐状态对所述城市坐标进行编码，得到与所述城市坐标对应的编码隐状态；及
    通过所述神经网络模型利用所述初始解码隐状态对所述城市坐标对应的编码隐状态进行解码，得到与所述城市坐标对应的解码隐状态。
  3. 根据权利要求2所述的方法，其特征在于，所述神经网络模型通过多个编码环节对城市坐标进行编码，所述通过所述神经网络模型利用所述初始编码隐状态对所述城市坐标进行编码，得到与所述城市坐标对应的编码隐状态，包括：
    通过所述神经网络模型在所述城市坐标中选取目标坐标；
    通过所述神经网络模型利用初始解码隐状态对所述目标坐标进行编码，输出与所述目标坐标对应的编码隐状态，将所述编码隐状态传送至下一个编码环节；
    通过所述神经网络模型根据所述编码隐状态对所述城市坐标中的下一个城市坐标进行编码，得到下一个编码隐状态；及
    重复通过所述神经网络模型利用所述初始编码隐状态对所述城市坐标进行编码的步骤，直至得到与所述城市坐标对应的编码隐状态。
  4. 根据权利要求3所述的方法，其特征在于，所述神经网络模型通过多个解码环节对城市坐标进行解码，所述通过所述神经网络模型利用所述初始解码隐状态对所述城市坐标对应的编码隐状态进行解码，得到与所述城市坐标对应的解码隐状态，包括：
    通过所述神经网络模型利用所述初始解码隐状态对所述目标坐标对应的编码隐状态 进行解码，得到所述目标坐标对应的解码隐状态，将所述解码隐状态传送至下一个解码环节；
    通过所述神经网络模型根据所述解码隐状态对所述下一个编码隐状态进行解码；及
    重复通过所述神经网络模型利用所述初始解码隐状态对所述城市坐标对应的编码隐状态进行解码的步骤，直至得到与所述城市坐标对应的解码隐状态。
  5. 根据权利要求1所述的方法，其特征在于，所述利用所述神经网络模型根据所述城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率，包括：
    将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进行比对，得到每个城市坐标对应的解码隐状态与多个城市标识对应的编码隐状态的相似度；及
    将所述相似度进行归一化处理，得到每个城市标识的路线概率。
  6. 根据权利要求5所述的方法，其特征在于，所述根据路线概率生成多个城市之间的最短路线序列，包括：
    利用所述解码隐状态与对应的相似度生成每个城市标识对应的注意力指针；及
    根据所述注意力指针指向所述路线概率中的最大概率，将所述最大概率对应的城市标识进行输出。
  7. 根据权利要求5所述的方法，其特征在于，所述方法还包括：
    将所述城市坐标对应的编码隐状态的位置信息进行存储；及
    根据所述位置信息将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进行比对，得到所述解码隐状态对应的相似度。
  8. 根据权利要求1至7任意一项所述的方法，其特征在于，所述方法还包括：
    根据所述城市坐标获取预设最短路线序列；及
    根据所述城市坐标、预设最短路线序列、最短路线序列以及预设关系计算所述最短路线序列的损失。
  9. 一种最短路线计算装置，包括：
    通信模块，用于获取路线计算任务，所述路线计算任务携带多个城市标识；
    坐标获取模块，用于根据所述城市标识获取对应的城市坐标；
    状态计算模块，用于根据所述路线计算任务调用神经网络模型，利用所述神经网络模型计算所述城市坐标对应的编码隐状态以及解码隐状态；
    概率计算模块，用于利用所述神经网络模型根据所述城市坐标对应的编码隐状态、解 码隐状态以及预设关系计算每个城市标识的路线概率；及
    序列生成模块，用于根据所述路线概率生成多个城市之间的最短路线序列。
  10. 根据权利要求9所述的装置，其特征在于，所述状态计算模块用于通过所述神经网络模型利用所述初始编码隐状态对所述城市坐标进行编码，得到与所述城市坐标对应的编码隐状态；及通过所述神经网络模型利用所述初始解码隐状态对所述城市坐标对应的编码隐状态进行解码，得到与所述城市坐标对应的解码隐状态。
  11. 一种计算机设备，包括存储器及一个或多个处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述一个或多个处理器执行时，使得所述一个或多个处理器执行以下步骤：
    使得所述一个或多个处理器执行以下步骤：
    获取路线计算任务，所述路线计算任务携带多个城市标识；
    根据所述城市标识获取对应的城市坐标；
    根据所述路线计算任务调用神经网络模型，利用所述神经网络模型计算所述城市坐标对应的编码隐状态以及解码隐状态；
    利用所述神经网络模型根据所述城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率；及
    根据所述路线概率生成多个城市之间的最短路线序列。
  12. 根据权利要求11所述的计算机设备，其特征在于，所述处理器执行所述计算机可读指令时还执行以下步骤：通过所述神经网络模型利用所述初始编码隐状态对所述城市坐标进行编码，得到与所述城市坐标对应的编码隐状态；及通过所述神经网络模型利用所述初始解码隐状态对所述城市坐标对应的编码隐状态进行解码，得到与所述城市坐标对应的解码隐状态。
  13. 根据权利要求12所述的计算机设备，其特征在于，所述处理器执行所述计算机可读指令时还执行以下步骤：通过所述神经网络模型在所述城市坐标中选取目标坐标；通过所述神经网络模型利用初始解码隐状态对所述目标坐标进行编码，输出与所述目标坐标对应的编码隐状态，将所述编码隐状态传送至下一个编码环节；通过所述神经网络模型根据所述编码隐状态对所述城市坐标中的下一个城市坐标进行编码，得到下一个编码隐状态；及重复通过所述神经网络模型利用所述初始编码隐状态对所述城市坐标进行编码的步骤，直至得到与所述城市坐标对应的编码隐状态。
  14. 根据权利要求12所述的计算机设备，其特征在于，所述处理器执行所述计算机 可读指令时还执行以下步骤：通过所述神经网络模型利用所述初始解码隐状态对所述目标坐标对应的编码隐状态进行解码，得到所述目标坐标对应的解码隐状态，将所述解码隐状态传送至下一个解码环节；通过所述神经网络模型根据所述解码隐状态对所述下一个编码隐状态进行解码；及重复通过所述神经网络模型利用所述初始解码隐状态对所述城市坐标对应的编码隐状态进行解码的步骤，直至得到与所述城市坐标对应的解码隐状态。
  15. 根据权利要求12所述的计算机设备，其特征在于，所述处理器执行所述计算机可读指令时还执行以下步骤：将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进行比对，得到每个城市坐标对应的解码隐状态与多个城市标识对应的编码隐状态的相似度；及将所述相似度进行归一化处理，得到每个城市标识的路线概率。
  16. 一个或多个存储有计算机可读指令的非易失性计算机可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行以下步骤：
    获取路线计算任务，所述路线计算任务携带多个城市标识；
    根据所述城市标识获取对应的城市坐标；
    根据所述路线计算任务调用神经网络模型，利用所述神经网络模型计算所述城市坐标对应的编码隐状态以及解码隐状态；
    利用所述神经网络模型根据所述城市坐标对应的编码隐状态、解码隐状态以及预设关系计算每个城市标识的路线概率；及
    根据所述路线概率生成多个城市之间的最短路线序列。
  17. 根据权利要求16所述的存储介质，其特征在于，所述计算机可读指令被所述处理器执行时还执行以下步骤：通过所述神经网络模型利用所述初始编码隐状态对所述城市坐标进行编码，得到与所述城市坐标对应的编码隐状态；及通过所述神经网络模型利用所述初始解码隐状态对所述城市坐标对应的编码隐状态进行解码，得到与所述城市坐标对应的解码隐状态。
  18. 根据权利要求17所述的存储介质，其特征在于，所述计算机可读指令被所述处理器执行时还执行以下步骤：通过所述神经网络模型在所述城市坐标中选取目标坐标；通过所述神经网络模型利用初始解码隐状态对所述目标坐标进行编码，输出与所述目标坐标对应的编码隐状态，将所述编码隐状态传送至下一个编码环节；通过所述神经网络模型根据所述编码隐状态对所述城市坐标中的下一个城市坐标进行编码，得到下一个编码隐状态；及重复通过所述神经网络模型利用所述初始编码隐状态对所述城市坐标进行编码的步骤，直至得到与所述城市坐标对应的编码隐状态。
  19. 根据权利要求17所述的存储介质，其特征在于，所述计算机可读指令被所述处理器执行时还执行以下步骤：通过所述神经网络模型利用所述初始解码隐状态对所述目标坐标对应的编码隐状态进行解码，得到所述目标坐标对应的解码隐状态，将所述解码隐状态传送至下一个解码环节；通过所述神经网络模型根据所述解码隐状态对所述下一个编码隐状态进行解码；及重复通过所述神经网络模型利用所述初始解码隐状态对所述城市坐标对应的编码隐状态进行解码的步骤，直至得到与所述城市坐标对应的解码隐状态。
  20. 根据权利要求16所述的存储介质，其特征在于，所述计算机可读指令被所述处理器执行时还执行以下步骤：将每个城市坐标对应的解码隐状态与多个城市坐标对应的编码隐状态进行比对，得到每个城市坐标对应的解码隐状态与多个城市标识对应的编码隐状态的相似度；及将所述相似度进行归一化处理，得到每个城市标识的路线概率。
PCT/CN2019/116647 2019-01-22 2019-11-08 最短路线计算方法、装置、计算机设备和存储介质 Ceased WO2020151309A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910056343.2 2019-01-22
CN201910056343.2A CN109829579B (zh) 2019-01-22 2019-01-22 最短路线计算方法、装置、计算机设备和存储介质

Publications (1)

Publication Number Publication Date
WO2020151309A1 true WO2020151309A1 (zh) 2020-07-30

Family

ID=66860458

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/116647 Ceased WO2020151309A1 (zh) 2019-01-22 2019-11-08 最短路线计算方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN109829579B (zh)
WO (1) WO2020151309A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109829579B (zh) * 2019-01-22 2023-01-24 平安科技（深圳）有限公司 最短路线计算方法、装置、计算机设备和存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102073669A (zh) * 2010-10-15 2011-05-25 江苏鸿信系统集成有限公司 基于全国范围内路网结构的任意城市间最优路径选择方法
US20150323337A1 (en) * 2014-05-07 2015-11-12 Yahoo! Inc. System and method for recommending pleasant routes from the sentiment of geo-tagged photographs
CN106991159A (zh) * 2017-03-30 2017-07-28 百度在线网络技术（北京）有限公司 用于生成信息的方法和装置
CN108303113A (zh) * 2018-01-26 2018-07-20 北京小度信息科技有限公司 导航路径优化方法、装置、电子设备及计算机存储介质
CN108628904A (zh) * 2017-03-23 2018-10-09 北京嘀嘀无限科技发展有限公司 一种路径编码、相似路径检索方法及装置和电子设备
CN109829579A (zh) * 2019-01-22 2019-05-31 平安科技（深圳）有限公司 最短路线计算方法、装置、计算机设备和存储介质

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10565318B2 (en) * 2017-04-14 2020-02-18 Salesforce.Com, Inc. Neural machine translation with latent tree attention
US20180330718A1 (en) * 2017-05-11 2018-11-15 Mitsubishi Electric Research Laboratories, Inc. System and Method for End-to-End speech recognition
CN107484017B (zh) * 2017-07-25 2020-05-26 天津大学 基于注意力模型的有监督视频摘要生成方法
CN109116100B (zh) * 2018-07-09 2019-09-27 清华大学 一种基于编码-解码结构的电力负荷用电分解方法
CN109241291B (zh) * 2018-07-18 2022-02-15 华南师范大学 基于深度强化学习的知识图谱最优路径查询系统及其方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102073669A (zh) * 2010-10-15 2011-05-25 江苏鸿信系统集成有限公司 基于全国范围内路网结构的任意城市间最优路径选择方法
US20150323337A1 (en) * 2014-05-07 2015-11-12 Yahoo! Inc. System and method for recommending pleasant routes from the sentiment of geo-tagged photographs
CN108628904A (zh) * 2017-03-23 2018-10-09 北京嘀嘀无限科技发展有限公司 一种路径编码、相似路径检索方法及装置和电子设备
CN106991159A (zh) * 2017-03-30 2017-07-28 百度在线网络技术（北京）有限公司 用于生成信息的方法和装置
CN108303113A (zh) * 2018-01-26 2018-07-20 北京小度信息科技有限公司 导航路径优化方法、装置、电子设备及计算机存储介质
CN109829579A (zh) * 2019-01-22 2019-05-31 平安科技（深圳）有限公司 最短路线计算方法、装置、计算机设备和存储介质

Also Published As

Publication number Publication date
CN109829579A (zh) 2019-05-31
CN109829579B (zh) 2023-01-24

Similar Documents

Publication Publication Date Title
CN111061847A (zh) 对话生成及语料扩充方法、装置、计算机设备和存储介质
WO2020253357A1 (zh) 数据产品推荐方法、装置、计算机设备和存储介质
CN109977635B (zh) 基于区块链的溯源数据处理方法、装置、设备及介质
CN113987154A (zh) 基于UniLM与对比学习的相似句生成模型训练方法及相关设备
WO2020252911A1 (zh) 失踪人脸识别方法、装置、计算机设备和存储介质
US11714921B2 (en) Image processing method with ash code on local feature vectors, image processing device and storage medium
CN110489622B (zh) 对象信息的分享方法、装置、计算机设备和存储介质
CN112084752B (zh) 基于自然语言的语句标注方法、装置、设备及存储介质
CN113343711B (zh) 工单生成方法、装置、设备及存储介质
CN111753318A (zh) 私有数据的多方安全计算方法、装置及系统
CN115329876A (zh) 一种设备故障处理方法、装置、计算机设备及存储介质
CN113222143A (zh) 图神经网络训练方法、系统、计算机设备及存储介质
WO2019075967A1 (zh) 企业名称识别方法、电子设备及计算机可读存储介质
CN112449009A (zh) 一种基于svd的联邦学习推荐系统通信压缩方法及装置
WO2022141870A1 (zh) 基于人工智能的语音合成方法、装置、计算机设备和介质
CN113220946B (zh) 基于强化学习的故障链路搜索方法、装置、设备及介质
CN114782237A (zh) 基于图案编码的水印生成方法、装置、设备及存储介质
WO2020151309A1 (zh) 最短路线计算方法、装置、计算机设备和存储介质
CN112732884A (zh) 目标回答语句生成方法、装置、计算机设备和存储介质
US20210117788A1 (en) Secure data processing
WO2023024287A1 (zh) Rpa结合ai的模型融合结果获取方法、装置及电子设备
CN116663677B (zh) 模型训练方法、装置、电子设备及存储介质
CN114358452B (zh) 用于预测竞争和合作关系的图模型处理方法、装置
CN113434652B (zh) 智能问答方法、智能问答装置、设备及存储介质
CN118051215A (zh) 程序生成方法、装置、计算机设备和存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19911951

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19911951

Country of ref document: EP

Kind code of ref document: A1