WO2021227535A1 - 预测节点状态的方法和装置 - Google Patents

预测节点状态的方法和装置 Download PDF

Info

Publication number
WO2021227535A1
WO2021227535A1 PCT/CN2020/141253 CN2020141253W WO2021227535A1 WO 2021227535 A1 WO2021227535 A1 WO 2021227535A1 CN 2020141253 W CN2020141253 W CN 2020141253W WO 2021227535 A1 WO2021227535 A1 WO 2021227535A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
period
graph
target
dynamic
Prior art date
Application number
PCT/CN2020/141253
Other languages
English (en)
French (fr)
Inventor
张成芝
周敏
蚁韩羚
马凯伦
庄克琛
Original Assignee
华为技术有限公司
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 华为技术有限公司 filed Critical 华为技术有限公司
Priority to EP20935951.2A priority Critical patent/EP4135264A4/en
Publication of WO2021227535A1 publication Critical patent/WO2021227535A1/zh
Priority to US17/984,421 priority patent/US20230117633A1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/147—Network analysis or design for predicting network behaviour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G06N5/022—Knowledge engineering; Knowledge acquisition
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/12—Discovery or management of network topologies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/02—Arrangements for optimising operational condition

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a method and device for predicting the state of a node.
  • the current state of a node is not only related to time, but also related to the space in which the node is located.
  • the traffic carried by a base station during the day is different from the traffic carried at night, that is, the traffic carried by the base station is related to time; in addition, the traffic carried by the base station at the center of the network is the same as the traffic carried by the base station at the edge of the network.
  • the traffic is different, that is, there is an association between the traffic carried by the base station and the space. In order to facilitate network management and maintenance, it is necessary to predict the state of nodes in the network for a period of time in the future.
  • a method of predicting the state of a node is to obtain a preset network topology map, and predict the state of a node in the network based on the preset network topology map and some network-related time information. This method is suitable for the topology structure in a long time For networks in a fixed state, for networks where the topological structure of a cellular network changes rapidly over time, the effect of applying the above method is not ideal.
  • the present application provides a method and device for predicting the state of a node, which can improve the prediction accuracy of the state of a node in a network with a dynamic topological relationship.
  • a method for predicting the state of a node including: obtaining a static graph and a dynamic graph of a plurality of nodes in a target network, the static graph and the dynamic graph are both topological graphs; according to the static graph and The dynamic graph generates the spatial feature data of the multiple nodes; obtains the temporal feature data of the multiple nodes; obtains the predicted state of the target node in the target period according to the spatial feature data and the temporal feature data ,
  • the target node is any one of the multiple nodes.
  • dynamic spatial features are also used, thereby improving the prediction accuracy of the node state in a network with dynamic topological relationships, and making the network in terms of node state prediction More intelligent.
  • the acquiring time characteristic data of the multiple nodes includes: acquiring observation data of the multiple nodes in a first historical period, where the first historical period is a period before the target period; The observation data is sliced in the time dimension to obtain time slice data, where the data amount of the time slice data is smaller than the data amount of the observation data; and the time characteristic data is obtained according to the time slice data.
  • Time slice data is part of the data selected from the complete observation data set. Compared with the complete observation data set, the data volume of time slice data is smaller. Using time slice data for prediction can reduce the time required for prediction.
  • the time slice data includes: observation data in a period adjacent to the target period in the first historical period.
  • the observation data in a specific period in the first historical period wherein the specific period is located in the first period, the target period is located in the second period, and the specific period includes the target period in the The corresponding time period in the first period.
  • the adjacent time period is close to the target time period, and the specific time period and the target time period are in the corresponding time domain positions of different periods. Therefore, the observation data in the adjacent time period and the specific time period have a strong correlation with the state of the node in the target time period. Use Predicting observation data in adjacent time periods and/or specific time periods is beneficial to improve prediction accuracy.
  • the obtaining static graphs and dynamic graphs of multiple nodes includes: obtaining observation data of the multiple nodes in a second historical period, where the second historical period is a period before the target period; According to the observation data, the static graph and the dynamic graph are constructed.
  • the second historical period and the first historical period may be the same or different. Use observation data to construct static maps and dynamic maps to obtain the latest spatial features in time.
  • the observation data includes at least one of the following data: weather data, network topology data, traffic data, voice data, signaling data, point of interest POI data, major event data, and holiday data.
  • Network topology data, traffic data, voice data, signaling data and POI data can be used to construct static graphs and dynamic graphs in order to obtain spatial characteristic data; weather data, major event data and holiday data can be used to obtain temporal characteristic data,
  • the aforementioned observation data used to obtain spatial feature data can also be used to obtain temporal feature data.
  • the above-mentioned data has different sources and different data structures. It can be called multi-source heterogeneous data. Obtaining temporal characteristic data and spatial characteristic data from multi-source heterogeneous data enables the temporal model and spatial model to learn more abundant knowledge. Improve forecast accuracy.
  • it further includes: acquiring the real state of the target node in the target period; training a temporal model and a spatial model according to the real state and the predicted state, wherein the temporal model is used to The observation data outputs the temporal feature data, and the spatial model is used to output the spatial feature data according to the input static image and the dynamic image.
  • the static graph belongs to the first type of topological graph
  • the dynamic graph belongs to the second type of topological graph
  • the topological relationship change rate of the first type of topological graph is less than the topological relationship change of the second type of topological graph rate
  • the target network is a communication network
  • the static graph is a topological graph that characterizes the topological relationship of the multiple nodes based on traffic
  • the dynamic graph is a topological graph that characterizes the topological relationship of the multiple nodes based on physical lines. Topology.
  • the present application provides an apparatus for predicting the state of a node, including a unit for executing the method described in the first aspect.
  • the device can be a terminal device or a server, or a chip in the terminal device or the server.
  • the device may include an input unit and a processing unit.
  • the processing unit may be a processor, and the input unit may be a transceiver; the terminal device may also include a storage unit, and the storage unit may be a memory; the storage unit is used to store instructions, The processing unit executes the instructions stored in the storage unit, so that the terminal device executes the method described in the first aspect.
  • the processing unit may be a processing unit inside the chip, and the input unit may be an input/output interface, a pin or a circuit, etc.; the processing unit executes instructions stored in the storage unit , So that the chip performs the method described in the first aspect, the storage unit can be a storage unit (for example, a register, a cache, etc.) in the chip, or a storage unit (for example, a read-only memory) located outside the chip. , Random Access Memory, etc.).
  • the present application provides a computer-readable storage medium in which a computer program is stored.
  • the processor executes the method described in the first aspect.
  • the present application provides a computer program product, the computer program product comprising: computer program code, when the computer program code is executed by a processor, the processor executes the method described in the first aspect.
  • Figure 1 shows a network suitable for this application
  • Figure 2 is a schematic diagram of a dynamic picture and a static picture provided by this application;
  • FIG. 3 is a schematic diagram of the method for predicting the state of a node provided by the present application.
  • Figure 4 is a schematic diagram of a historical traffic provided by this application.
  • Fig. 5 is a schematic diagram of a method for acquiring time slice data provided by the present application.
  • Fig. 6 is a schematic diagram of a method for predicting the state of a base station provided by the present application.
  • FIG. 7 is a schematic diagram of a device for predicting the state of a node provided by the present application.
  • Fig. 8 is a schematic diagram of a device for predicting the state of a node provided by the present application.
  • FIG. 1 shows a network suitable for this application.
  • the network 100 includes a plurality of base stations, and the base stations may be referred to as nodes.
  • the traffic between the base stations and the base stations together form a topology of the network 100, and the physical lines between the base stations and the base stations share the same topology.
  • Another topology diagram that constitutes the network 100 Affected by the user's usage behavior and mobile behavior, the traffic carried by each base station has a strong correlation with the time dimension and the space dimension. Therefore, the topological graph based on the traffic keeps changing over time, and this topological relationship changes rapidly over time The topological graph can be called a dynamic graph.
  • the physical line is not affected by the user's usage behavior and mobile behavior.
  • the topological relationship of this topology map changes slowly over time, so it can be called a static map.
  • the dynamic graph and static graph of the network 100 are shown in FIG. 2.
  • the black dots represent the base station.
  • the lines between the black dots represent the traffic; in the static graph, the lines between the black dots represent the physical lines.
  • substations can be regarded as nodes.
  • the transmission lines and substations between substations together constitute a static diagram of the grid, and the currents between substations and the substation together constitute a dynamic diagram of the grid.
  • users can be regarded as nodes.
  • the relationship between friends and users and these users together constitute a static graph of the social network, and the relationship between unfamiliar users and these users together constitute a dynamic graph of the social network.
  • the following takes the network 100 as an example to introduce the technical solution of the present application in detail.
  • the method 300 includes:
  • S310 Acquire static graphs and dynamic graphs of multiple nodes in the target network, where both the static graph and the dynamic graph are topological graphs.
  • Static graphs and dynamic graphs belong to undirected and unauthorised graphs, which can be passed Represents an undirected unweighted graph, where v represents a collection of nodes, and ⁇ represents a collection of connections between nodes.
  • the device that executes the method 300 can construct connection matrices corresponding to different graphs through a variety of methods.
  • the distance between the base stations can be determined, and then the connection matrix between the base stations can be determined according to formula (1), and then the static image can be determined.
  • a SD (i,j) represents the connection relationship between base station i and base station j determined according to the distance.
  • a SD (i, j) 0, it indicates the absence of a connection relationship between the base station i and station j;
  • d ij represents the distance between the base station i and station j;
  • e represents a natural constant; ultra parameter [sigma] is used to control the base station from The magnitude of the value; Is a mapping function used to Mapped to 1 or 0.
  • the similarity can be calculated from the historical traffic, and then the connection matrix between the base stations is determined according to formula (2), and then the static graph is determined.
  • a FS (i,j) represents the connection relationship between base station i and base station j determined according to historical traffic.
  • a When FS (i, j) 0, it means that there is no connection relationship between base station i and base station j
  • x w (i) represents the weekly average data of base station i's traffic
  • x w (j) represents the weekly traffic of base station j
  • Average data Represents the Pearson correlation coefficient of x w (i) and x w (j); Is a mapping function used to Mapped to 1 or 0.
  • a static graph can be used as the initial graph, and then the initial graph can be updated with real-time data to obtain a dynamic graph.
  • real-time data can be used to directly construct a dynamic graph, as shown in formula (3).
  • x t (i) (x t-H+1 (i),...,x t-1 (i)), which means The set of traffic of base station i in the H time closest to time t;
  • x t (j) (x t-H+1 (j),...,x t-1 (j)), which means that The set of traffic of base station j in the H most recent moments at time t;
  • mapping variables is shown in formula (4).
  • z means or or ⁇ is a hyperparameter, which is used to control the sparseness of the topological graph.
  • the method for determining the static image and the dynamic image applicable to the present application is not limited to the above examples.
  • the apparatus for executing the method 300 can execute the following steps.
  • S320 Generate spatial feature data of the multiple nodes according to the static graph and the dynamic graph.
  • GCN graph neural networks
  • a convolution kernel of GCN can be used to perform convolution operations on the static images and dynamic images to obtain the number of spatial features.
  • S320 and S330 are executed in no particular order.
  • the observation data of multiple nodes in a historical period can be processed by a gate recurrent unit (GRU) in a recursive neural network (RNN) to obtain time characteristic data.
  • GRU gate recurrent unit
  • RNN recursive neural network
  • the aforementioned observation data includes at least one of the following data: weather data, network topology data, traffic data, voice data, signaling data, point of interest (POI) data, major event data, and holiday data.
  • the amount of observation data is large, and it takes longer to directly use the observation data to make predictions. Therefore, before processing the observation data, the observation data can be preprocessed to reduce the time required for prediction.
  • the observation data can be sliced (ie, sampled) by using characteristics such as temporal proximity and periodicity of the observation data, and time slice data can be determined from the observation data, where the data amount of the time slice data is smaller than the data amount of the observation data; Then the time feature data is obtained by processing the time slice data through GRU.
  • Figure 4 shows the traffic data in the historical period (20 days) of the target base station. It can be seen that the traffic data exhibits obvious periodicity.
  • the flow data can be sliced based on the position of the target time period (that is, the time period to be predicted) within one period, and the flow data in the time period corresponding to the target time period in other periods can be determined as time slice data, because the corresponding time period in different periods.
  • the data within is similar, and prediction based on the time slice data obtained by this feature can reduce the time required for prediction without affecting the prediction effect.
  • Figure 5 shows a method of acquiring time slice data.
  • the target time period is 8:15-9:15 in the morning (am) of September 1, 2018 (Sat.). It can be a day (day) or a week (week) cycle.
  • the flow data in the time period corresponding to the target time period is determined as time slice data.
  • the target period (8:15-9:15 in the morning) is in the second cycle (September 1, 2018)
  • August 30, 2018 can be used as the first cycle
  • August 30, 2018 8:15-9:15 in the morning is the time period corresponding to the target time period in the first cycle.
  • the time period that includes the corresponding time period can be called a specific time period.
  • the specific time period on August 30, 2018 is 6:30-9:30 am.
  • it can also be determined that the 3:1-9:30 am on August 30, 2018 is a specific time period.
  • each specific period can be referred to as T d .
  • the 26th can be used as the first cycle.
  • 8:15-9:15 in the morning of August 25, 2018 (Saturday) is the time period corresponding to the target time period in the first cycle.
  • the time period that contains the corresponding time period can be called a specific time period.
  • Time period in the example in FIG. 5, the specific time period from August 20, 2018 to August 26, 2018 is 6:30-9:30 in the morning on August 25, 2018 (Saturday).
  • it can also be determined that the 3:1-9:30 am on August 11, 2018 (Saturday) is a specific time period.
  • each specific period can be referred to as T w .
  • the observation data within a specific time period is the time slice data.
  • time slice data can also be determined from the traffic data in the historical period with a month or quarter cycle.
  • the time slice data also includes traffic data in a time period adjacent to the target time period.
  • the adjacent time period is, for example, 5:00-8:00 am on September 1, 2018, and 2018 the afternoon of August 30 (pm) 8: 00-11: 00 , in this example, each adjacent period can be called T r.
  • time slice data can be used independently or in combination.
  • T 3 can be selected D, 3, and 4 T w data within a time T r as the slice data, the time data of a total of 120 slice data (data corresponding to 8 hours).
  • the 20-day traffic data has a total of about 1920 data. Compared with time slice data, using time slice data for prediction can reduce the time required for prediction without affecting the prediction effect.
  • the apparatus for executing the method 300 can execute the following steps.
  • Figure 6 shows a prediction method provided by this application.
  • X t represents the data obtained after time slicing processing of the base station observation data before time t
  • X t is input to a temporal model (temporal modeling), and after the processing of the temporal model, the temporal characteristic data Y t is output.
  • the temporal model is, for example, It's GRU.
  • the static image and the dynamic image are input into the spatial modeling, and the spatial feature data is output through the processing of the spatial model.
  • the spatial model is, for example, GCN
  • the temporal feature data includes Indicates the use of GCN's convolution kernel ⁇ in the figure
  • the result of the convolution operation on It is one of the dynamic picture and the static picture before time t.
  • ReLU rectified linear unit
  • the final output result Z t of the space model can be obtained, as shown in formula (6).
  • Z t can be processed through a fully connected (FC) layer to obtain As shown in formula (7).
  • W f is the weight parameter of the FC layer
  • b f is the bias parameter of the FC layer
  • the time model, the space model, and the FC layer can be jointly trained by minimizing the loss function.
  • An optional loss function is shown in formula (8).
  • x t+k is the true state of the target base station at time t+k
  • ⁇ is used to balance the weight of MSE and MAD
  • N is the number of nodes.
  • the methods involved in the comparison include historical average (HA), prophetic model (Prophet), spatio-temporal graph convolutional network (STGCN), attention spatio-temporal graph convolutional network (attention spatio-temporal graph convolutional) network, ASTGCN) and diffusion convolutional recurrent neural network (diffusion convolutional recurrent neural network, DCRNN), the evaluation indicators are root mean square error (RMSE) and MAE, the smaller the value of RMSE and MAE, the smaller the value of the method is The better the predictive ability.
  • HA historical average
  • Praphet prophetic model
  • STGCN spatio-temporal graph convolutional network
  • attention spatio-temporal graph convolutional network attention spatio-temporal graph convolutional network
  • ASTGCN attention spatio-temporal graph convolutional network
  • DCRNN diffusion convolutional recurrent neural network
  • the comparison results are shown in Table 1.
  • STGCN, ASTGCN, DCRNN, and method 300 are all methods based on deep learning, and the results of each run will have some differences. Table 1 shows the results of these methods for 10 predictions, and gives the mean and variance. It can be seen from Table 1 that the method 300 has achieved the best prediction effect among several methods.
  • the corresponding device includes a hardware structure and/or software module corresponding to each function.
  • the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed by hardware or computer software-driven hardware depends on the specific application and design constraint conditions of the technical solution. Professionals and technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered beyond the scope of this application.
  • the present application may divide the device into functional units according to the foregoing method examples. For example, each function may be divided into each functional unit, or two or more functions may be integrated into one unit.
  • the above-mentioned integrated unit can be implemented in the form of hardware or software functional unit. It should be noted that the division of units in this application is illustrative, only a logical function division, and other division methods may be used in actual implementation.
  • Fig. 7 shows a schematic structural diagram of a device for predicting the state of a node provided by the present application.
  • the device 700 includes an input unit 710 and a processing unit 720.
  • the processing unit 720 is configured to: obtain static graphs and dynamic graphs of multiple nodes in the target network, where the static graphs and the dynamic graphs are both topological graphs; according to the static graphs and the dynamic graphs, generate the multiple The spatial characteristic data of the node; acquiring the temporal characteristic data of the multiple nodes; acquiring the predicted state of the target node in the target period according to the spatial characteristic data and the temporal characteristic data, and the target node is the multiple Any one of the nodes.
  • the input unit 710 is configured to: obtain observation data of the multiple nodes in a first historical period, where the first historical period is a period before the target period; the processing unit 720 is specifically configured to: The observation data is sliced dimensionally to obtain time slice data, where the data amount of the time slice data is smaller than the data amount of the observation data; and the time characteristic data is obtained according to the time slice data.
  • the time slice data includes: observation data in a period adjacent to the target period in the first historical period.
  • the time slice data includes: observation data in a specific period in the first historical period; wherein, the specific period is located in the first period, the target period is located in the second period, and the specific period is The period includes a period corresponding to the target period in the first period.
  • the input unit 710 is configured to: obtain observation data of the multiple nodes in a second historical period, where the second historical period is a period before the target period; the processing unit 720 is specifically configured to: The observation data, the static graph and the dynamic graph are constructed.
  • the observation data includes at least one of the following data: weather data, network topology data, traffic data, voice data, signaling data, POI data, major event data, and holiday data.
  • the processing unit 720 is further configured to: obtain the real state of the target node in the target period; train a time model and a space model according to the real state and the predicted state, wherein the time model is To output the temporal feature data according to the input observation data, the spatial model is used to output the spatial feature data according to the input static image and the dynamic image.
  • the static graph belongs to the first type of topological graph
  • the dynamic graph belongs to the second type of topological graph
  • the topological relationship change rate of the first type of topological graph is less than the topological relationship change of the second type of topological graph rate
  • the target network is a communication network
  • the static graph is a topological graph that characterizes the topological relationship of the multiple nodes based on traffic
  • the dynamic graph is a topological graph that characterizes the topological relationship of the multiple nodes based on physical lines. Topology.
  • Fig. 8 shows a schematic structural diagram of a device for predicting the state of a node provided by the present application.
  • the dotted line in Figure 8 indicates that the unit or the module is optional.
  • the device 800 may be used to implement the methods described in the foregoing method embodiments.
  • the device 800 may be a terminal device or a server or a chip.
  • the device 800 includes one or more processors 801, and the one or more processors 801 can support the device 800 to implement the method in the method embodiment.
  • the processor 801 may be a general-purpose processor or a special-purpose processor.
  • the processor 801 may be a central processing unit (CPU).
  • the CPU can be used to control the device 800, execute a software program, and process data of the software program.
  • the device 800 may further include a communication unit 805 for implementing input (reception) and/or output (transmission) of signals (such as observation data and prediction results).
  • the device 800 may be a chip, and the communication unit 805 may be an input and/or output circuit of the chip, or the communication unit 805 may be a communication interface of the chip, and the chip may be used as a terminal device or a network device or other electronic device. component.
  • the device 800 may be a terminal device or a server
  • the communication unit 805 may be a transceiver of the terminal device or the server, or the communication unit 805 may be a transceiver circuit of the terminal device or the server.
  • the device 800 may include one or more memories 802 with a program 804 stored thereon, and the program 804 may be run by the processor 801 to generate instructions 803 so that the processor 801 executes the methods described in the foregoing method embodiments according to the instructions 803.
  • the memory 802 may also store data (such as observation data).
  • the processor 801 may also read data stored in the memory 802. The data may be stored at the same storage address as the program 804, or the data may be stored at a different storage address from the program 804.
  • the processor 801 and the memory 802 may be provided separately or integrated together, for example, integrated on a system-on-chip (SOC) of the terminal device.
  • SOC system-on-chip
  • the processor 801 may be a CPU, a digital signal processor (digital signal processor, DSP), an application specific integrated circuit (ASIC), a field programmable gate array (field programmable gate array, FPGA) or other programmable logic devices , For example, discrete gates, transistor logic devices, or discrete hardware components.
  • DSP digital signal processor
  • ASIC application specific integrated circuit
  • FPGA field programmable gate array
  • This application also provides a computer program product, which, when executed by the processor 801, implements the method described in any method embodiment in this application.
  • the computer program product may be stored in the memory 802, for example, a program 804, and the program 804 is finally converted into an executable object file that can be executed by the processor 801 through processing processes such as preprocessing, compilation, assembly, and linking.
  • the present application also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a computer, the method described in any method embodiment in the present application is implemented.
  • the computer program can be a high-level language program or an executable target program.
  • the computer-readable storage medium is, for example, the memory 802.
  • the memory 802 may be a volatile memory or a non-volatile memory, or the memory 802 may include both a volatile memory and a non-volatile memory.
  • the non-volatile memory can be read-only memory (ROM), programmable read-only memory (programmable ROM, PROM), erasable programmable read-only memory (erasable PROM, EPROM), and electrically available Erase programmable read-only memory (electrically EPROM, EEPROM) or flash memory.
  • the volatile memory may be random access memory (RAM), which is used as an external cache.
  • RAM random access memory
  • static random access memory static random access memory
  • dynamic RAM dynamic RAM
  • DRAM dynamic random access memory
  • synchronous dynamic random access memory synchronous DRAM, SDRAM
  • double data rate synchronous dynamic random access memory double data rate SDRAM, DDR SDRAM
  • enhanced synchronous dynamic random access memory enhanced SDRAM, ESDRAM
  • synchronous connection dynamic random access memory serial DRAM, SLDRAM
  • direct rambus RAM direct rambus RAM, DR RAM
  • the disclosed system, device, and method may be implemented in other ways. For example, some features of the method embodiments described above may be ignored or not implemented.
  • the device embodiments described above are merely illustrative.
  • the division of units is only a logical function division. In actual implementation, there may be other division methods, and multiple units or components may be combined or integrated into another system.
  • the coupling between the units or the coupling between the components may be direct coupling or indirect coupling, and the foregoing coupling includes electrical, mechanical, or other forms of connection.
  • the size of the sequence number does not mean the order of execution.
  • the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of this application. .

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • General Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请提供了一种预测节点状态的方法，包括：获取目标网络中多个节点的静态图和动态图，所述静态图和所述动态图均为拓扑图；根据所述静态图和所述动态图，生成所述多个节点的空间特征数据；获取所述多个节点的时间特征数据；根据所述空间特征数据和所述时间特征数据，获取目标节点在目标时段内的预测状态，所述目标节点为所述多个节点中的任意一个节点。本申请提供的预测节点状态的方法应用在在预测网络中节点的状态领域，除了使用了时间特征和静态空间特征，还使用了动态空间特征，从而提高了具有动态拓扑关系的网络中的节点状态的预测准确度，使得网络在节点状态预测方面更智能化。

Description

预测节点状态的方法和装置
本申请要求于2020年05月14日提交中国专利局、申请号为202010409392.2、申请名称为“预测节点状态的方法和装置”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
本申请涉及人工智能技术领域，具体涉及一种预测节点状态的方法和装置。
背景技术
在一些网络中，节点的当前状态除了与时间相关，还与该节点所处的空间相关。例如，在蜂窝网络中，一个基站在白天承载的流量与晚上承载流量不同，即，基站承载的流量与时间存在关联关系；此外，位于网络中心的基站承载的流量与位于网络边缘的基站承载的流量不同，即，基站承载的流量与空间存在关联关系。为了便于网络管理和维护，需要对网络中的节点在未来一段时间的状态进行预测。
一种预测节点状态的方法是获取预设的网络拓扑图，基于预设的网络拓扑图以及一些与网络相关的时间信息预测网络中节点的状态，这种方法适用于拓扑结构在较长时间内处于固定状态的网络，对于蜂窝网络等拓扑结构随时间发生较快变化的网络，应用上述方法的效果不够理想。
发明内容
本申请提供了一种预测节点状态的方法和装置，能够提高具有动态拓扑关系的网络中节点状态的预测准确度。
第一方面，提供了一种预测节点状态的方法，包括：获取目标网络中多个节点的静态图和动态图，所述静态图和所述动态图均为拓扑图；根据所述静态图和所述动态图，生成所述多个节点的空间特征数据；获取所述多个节点的时间特征数据；根据所述空间特征数据和所述时间特征数据，获取目标节点在目标时段内的预测状态，所述目标节点为所述多个节点中的任意一个节点。
在预测节点状态的过程中，除了使用了时间特征和静态空间特征，还使用了动态空间特征，从而提高了具有动态拓扑关系的网络中的节点状态的预测准确度，使得网络在节点状态预测方面更智能化。
可选地，所述获取所述多个节点的时间特征数据，包括：获取所述多个节点在第一历史时段内的观测数据，所述第一历史时段为所述目标时段之前的时段；在时间维度上对所述观测数据进行切片，获取时间切片数据，所述时间切片数据的数据量小于所述观测数据的数据量；根据所述时间切片数据获取所述时间特征数据。
时间切片数据是从完整的观测数据集中选取的部分数据，与完整的观测数据集相比， 时间切片数据的数据量较小，使用时间切片数据进行预测能够减少预测所需的时间。
可选地，所述时间切片数据包括：所述第一历史时段中与所述目标时段相邻的时段内的观测数据。
可选地，所述第一历史时段中特定时段内的观测数据；其中，所述特定时段位于第一周期内，所述目标时段位于第二周期内，所述特定时段包括所述目标时段在所述第一周期内对应的时段。
相邻时段与目标时段较近，特定时段与目标时段处于不同周期的对应时域位置，因此，相邻时段和特定时段内的观测数据与目标时段内节点的状态有较强的关联性，使用相邻时段和/或特定时段内的观测数据进行预测有利于提高预测准确度。
可选地，所述获取多个节点的静态图和动态图，包括：获取所述多个节点在第二历史时段内的观测数据，所述第二历史时段为所述目标时段之前的时段；根据所述观测数据，构建所述静态图和所述动态图。
第二历史时段与第一历史时段可以相同，也可以不同。利用观测数据构建静态图和动态图，能够及时获最新的取空间特征。
可选地，所述观测数据包括以下数据中的至少一个：气象数据、网络拓扑数据、流量数据、语音数据、信令数据、兴趣点POI数据、重大事件数据、以及节假日数据。
网络拓扑数据、流量数据、语音数据、信令数据和POI数据可以用于构建静态图和动态图，以便于获取空间特征数据；气象数据、重大事件数据和节假日数据可以用于获取时间特征数据，上述用于获取空间特征数据的观测数据也可以用于获取时间特征数据。上述数据的来源不同，数据结构也不同，可以称为多源异构数据，从多源异构数据中获取时间特征数据和空间特征数据，使得时间模型和空间模型能够学习到更丰富的知识，提高预测准确度。
可选地，还包括：获取所述目标节点在所述目标时段内的真实状态；根据所述真实状态和所述预测状态训练时间模型和空间模型，其中，所述时间模型用于根据输入的观测数据输出所述时间特征数据，所述空间模型用于根据输入的所述静态图和所述动态图输出所述空间特征数据。
通过训练，能够提高空间模型和时间模型的预测准确度。
可选地，所述静态图属于第一类拓扑图，所述动态图属于第二类拓扑图，所述第一类拓扑图的拓扑关系变化速率小于所述第二类拓扑图的拓扑关系变化速率。
可选地，所述目标网络为通信网络，所述静态图为表征所述多个节点基于流量的拓扑关系的拓扑图，所述动态图为表征所述多个节点基于物理线路的拓扑关系的拓扑图。
第二方面，本申请提供了一种预测节点状态的装置，包括用于执行第一方面所述的方法的单元。该装置可以是终端设备或服务器，也可以是终端设备或服务器内的芯片。该装置可以包括输入单元和处理单元。
当该装置是终端设备或服务器时，该处理单元可以是处理器，该输入单元可以是收发器；该终端设备还可以包括存储单元，该存储单元可以是存储器；该存储单元用于存储指令，该处理单元执行该存储单元所存储的指令，以使该终端设备执行第一方面所述的方法。
当该装置是终端设备或服务器内的芯片时，该处理单元可以是芯片内部的处理单元，该输入单元可以是输入/输出接口、管脚或电路等；该处理单元执行存储单元所存储的指 令，以使该芯片执行第一方面所述的方法，该存储单元可以是该芯片内的存储单元(例如，寄存器、缓存等)，也可以是位于该芯片外部的存储单元(例如，只读存储器、随机存取存储器等)。
第三方面，本申请提供了一种计算机可读存储介质，该计算机可读存储介质中存储了计算机程序，该计算机程序被处理器执行时，使得处理器执行第一方面所述的方法。
第四方面，本申请提供了一种计算机程序产品，该计算机程序产品包括：计算机程序代码，当该计算机程序代码被处理器运行时，使得处理器执行第一方面所述的方法。
附图说明
图1是示出了适用于本申请的一种网络；
图2是本申请提供的一种动态图和静态图的示意图；
图3是本申请提供的预测节点状态的方法的示意图；
图4是本申请提供的一种历史流量的示意图；
图5是本申请提供的一种获取时间切片数据的方法的示意图；
图6是本申请提供的一种预测基站状态的方法的示意图；
图7是本申请提供的一种预测节点状态的装置的示意图；
图8是本申请提供的一种预测节点状态的设备的示意图。
具体实施方式
下面将结合附图，对本申请的技术方案进行描述。
图1示出了适用于本申请的一种网络。网络100包括多个基站，基站可以称为节点。基站之间存在两种连接关系，一种是物理线路，另一种是流量，其中，基站之间的流量和基站共同构成了网络100的一种拓扑图，基站之间的物理线路和基站共同构成网络100的另一种拓扑图。受用户的使用行为和移动行为的影响，各个基站承载的流量与时间维度和空间维度具有较强的关联性，因此，基于流量的拓扑图随时间不断变化，这种拓扑关系随时间变化较快的拓扑图可以称为动态图。物理线路不受用户的使用行为和移动行为的影响，这种拓扑图的拓扑关系随时间变化较慢，因此，可以称为静态图。
网络100的动态图和静态图如图2所示。黑色的点表示基站，在动态图中，黑色的点之间的连线表示流量；在静态图中，黑色的点之间的连线表示物理线路。
除网络100之外，其它存在动态图和静态图的网络均适用于本申请。
例如，在电网中，变电站可以视为节点，变电站之间的输电线路和变电站共同构成电网的静态图，变电站之间的电流和变电站共同构成电网的动态图。
又例如，在社交网络中，用户可以视为节点，朋友用户之间的关系和这些用户共同构成社交网络的静态图，陌生用户之间的关系和这些用户共同构成社交网络的动态图。
下面以网络100为例详细介绍本申请的技术方案。
如图3所示，方法300包括：
S310，获取目标网络中多个节点的静态图和动态图，所述静态图和所述动态图均为拓扑图。
静态图和动态图属于无向无权图，可以通过
Figure PCTCN2020141253-appb-000001
表示一个无向无权图，其中，v 表示节点的集合，ε表示节点之间的连线的集合。可以通过连接矩阵A(i,j)表示节点i和节点j之间的连接关系，当A(i,j)＝1时，表示节点i和节点j之间存在连接关系；当A(i,j)＝0时，表示节点i和节点j之间不存在连接关系。
执行方法300的装置可以通过多种方法构建不同的图对应的连接矩阵。
例如，当基站的经纬度已知时，可以确定基站之间的距离，随后可以根据公式(1)确定基站之间的连接矩阵，进而确定静态图。
Figure PCTCN2020141253-appb-000002
A SD(i,j)表示根据距离确定的基站i和基站j之间的连接关系，当A SD(i,j)＝1时，表示基站i和基站j之间存在连接关系，当A SD(i,j)＝0时，表示基站i和基站j之间不存在连接关系；d ij表示基站i和基站j之间的距离；e表示自然常数；σ为超参数，用于控制基站距离数值的大小；
Figure PCTCN2020141253-appb-000003
是一个映射函数，用于将
Figure PCTCN2020141253-appb-000004
映射为1或0。
又例如，当基站的历史流量已知时，可以通过历史流量计算相似性，随后根据公式(2)确定基站之间的连接矩阵，进而确定静态图。
Figure PCTCN2020141253-appb-000005
A FS(i,j)表示根据历史流量确定的基站i和基站j之间的连接关系，当A FS(i,j)＝1时，表示基站i和基站j之间存在连接关系，当A FS(i,j)＝0时，表示基站i和基站j之间不存在连接关系；x w(i)表示基站i的流量的周平均数据；x w(j)表示基站j的流量的周平均数据；
Figure PCTCN2020141253-appb-000006
表示x w(i)和x w(j)的皮尔逊(pearson)相关系数；
Figure PCTCN2020141253-appb-000007
是一个映射函数，用于将
Figure PCTCN2020141253-appb-000008
映射为1或0。
上文介绍了确定静态图的方法，下面介绍确定动态图的方法。
例如，可以将一个静态图作为初始图，然后利用实时数据更新该初始图，从而获得动态图。
又例如，可以利用实时数据直接构造动态图，如公式(3)所示。
Figure PCTCN2020141253-appb-000009
Figure PCTCN2020141253-appb-000010
表示基于流量确定的基站i和基站j之间在时刻t的连接关系，x t(i)＝(x t-H+1(i),...,x t-1(i))，表示在离时刻t最近的H个时刻内基站i的流量的集合；x t(j)＝(x t-H+1(j),...,x t-1(j))，表示在离时刻t最近的H个时刻内基站j的流量的集合；
Figure PCTCN2020141253-appb-000011
表示x t(i)和x t(j)的皮尔逊相关系数；
Figure PCTCN2020141253-appb-000012
是一个映射函数，用于将
Figure PCTCN2020141253-appb-000013
映射为1或0。
可选地，上述几个示例中的
Figure PCTCN2020141253-appb-000014
映射变量的方法如公式(4)所示。
Figure PCTCN2020141253-appb-000015
z表示
Figure PCTCN2020141253-appb-000016
或
Figure PCTCN2020141253-appb-000017
或
Figure PCTCN2020141253-appb-000018
∈为超参数，用于控制拓扑图的疏密度。
适用于本申请的确定静态图和动态图的方法不限于上述示例。
获取静态图和动态图后，执行方法300的装置可以执行下列步骤。
S320，根据所述静态图和所述动态图，生成所述多个节点的空间特征数据。
例如，可以通过图神经网络(graph neural networks，GCN)处理静态图和动态图，使用GCN的卷积核在静态图和动态图上分别进行卷积操作，获取空间特征数。
S330，获取所述多个节点的时间特征数据。
S320和S330的执行顺序不分先后。可以通过递归神经网络(recursive neural network，RNN)中的门控循环单元(gate recurrent unit，GRU)处理多个节点在历史时段内的观测数据，获得时间特征数据。
上述观测数据包括以下数据中的至少一个：气象数据、网络拓扑数据、流量数据、语音数据、信令数据、兴趣点(point of interest，POI)数据、重大事件数据、以及节假日数据。
通常情况下，观测数据的数据量较大，直接使用观测数据进行预测所需的时间较长，因此，在处理观测数据之前，可以对观测数据进行预处理以减小预测所需的时间。
例如，可以利用观测数据的时间近邻性、周期性等特点对观测数据进行切片(即，采样)，从观测数据中确定时间切片数据，其中，时间切片数据的数据量小于观测数据的数据量；随后通过GRU处理时间切片数据获取时间特征数据。
下面介绍本申请提供的一种对流量数据进行时间切片的方法。
图4示出了目标基站的历史时段(20天)内的流量数据。可以看出，流量数据呈现明显的周期性。可以基于目标时段(即，待预测的时段)在一个周期内的位置对流量数据进行切片，将其它周期内与目标时段对应的时段内的流量数据确定为时间切片数据，由于不同周期内对应时段内的数据具有相似性，基于此特性获取的时间切片数据进行预测能够在不影响预测效果的同时减小预测所需的时间。
图5示出了一种获取时间切片数据的方法。
目标时段为2018年9月1日(周六(Sat.))的上午(a.m.)8:15-9:15，可以以天(day)或周(week)为周期，将相邻周期内与目标时段对应的时段内的流量数据确定为时间切片数据。
例如，目标时段(上午8:15-9:15)所在的周期(2018年9月1日)为第二周期，则2018年8月30日可以作为第一周期，2018年8月30日的上午8:15-9:15为目标时段在第一周期内对应的时段，可以将包含该对应时段的时段称为特定时段，在图5的示例中，2018年8月30日的特定时段为上午6:30-9:30。类似地，还可以确定2018年8月30日的上午6:30-9:30为特定时段。本示例中，每个特定时段可以称为T d。
又例如，目标时段(上午8:15-9:15)所在的周期(2018年8月27日至2018年9月2日)为第二周期，则2018年8月20日至2018年8月26日可以作为第一周期，2018年8月25日(周六)的上午8:15-9:15为目标时段在第一周期内对应的时段，可以将包含该对应时段的时段称为特定时段，在图5的示例中，2018年8月20日至2018年8月26日的特定时段为2018年8月25日(周六)的上午6:30-9:30。类似地，还可以确定2018年8月11日(周六)的上午6:30-9:30为特定时段。本示例中，每个特定时段可以称为T w。
确定特定时段后，特定时段内的观测数据即时间切片数据。
类似地，还可以以月(month)或季度(quarter)为周期从历史时段内的流量数据中确定时间切片数据。
图5所示的例子中，时间切片数据还包括与目标时段相邻的时段内的流量数据，该相邻的时段例如是2018年9月1日的上午5:00-8:00以及2018年8月30日的下午(p.m.)8:00-11:00，本示例中，每个相邻的时段可以称为T r。
需要说明的是，上述几个关于时间切片数据的示例可以独立使用，也可以结合使用。
若历史时段内的流量数据为间隔15分钟的数据，则图5中每个时间切片(T d、T w或T r)包含12个数据；可以选择3个T d、3个T w以及4个T r内的数据作为时间切片数据，则时间切片数据共有120个数据(相当于8个小时的数据)。20天的流量数据一共约1920个数据，与时间切片数据相比，使用时间切片数据进行预测能够在不影响预测效果的同时减小预测所需的时间。
获取时间特征数据和空间特征数据后，执行方法300的装置可以执行下列步骤。
S340，根据所述空间特征数据和所述时间特征数据，获取目标节点在目标时段内的预测状态，所述目标节点为所述多个节点中的任意一个节点。
图6示出了本申请提供的一种预测方法。
图6中，X t表示对t时刻之前的基站观测数据进行时间切片处理后得到的数据，X t输入时间模型(temporal modeling)，经过时间模型的处理输出时间特征数据Y t，该时间模型例如是GRU。
将静态图和动态图输入空间模型(spatial modeling)，经过空间模型的处理输出空间特征数据。空间模型例如是GCN，时间特征数据包括
Figure PCTCN2020141253-appb-000019
表示利用GCN的卷积核Θ在图
Figure PCTCN2020141253-appb-000020
上进行卷积操作的结果，
Figure PCTCN2020141253-appb-000021
为t时刻之前的动态图和静态图中的一个。
随后，将Y t输入空间模型，基于Y t和
Figure PCTCN2020141253-appb-000022
可以得到融合了时间特征和空间特征的结果
Figure PCTCN2020141253-appb-000023
如公式(5)所示。
Figure PCTCN2020141253-appb-000024
其中，ReLU表示线性整流函数(rectified linear unit)。
将多个
Figure PCTCN2020141253-appb-000025
累加，可以得到空间模型的最终输出结果Z t，如公式(6)所示。
Figure PCTCN2020141253-appb-000026
随后，可以通过一个全连接(fully connected，FC)层处理Z t，得到
Figure PCTCN2020141253-appb-000027
如公式(7)所示。
Figure PCTCN2020141253-appb-000028
其中，W f是FC层的权重参数，b f是FC层的偏置参数，
Figure PCTCN2020141253-appb-000029
表示目标基站在t+k时刻的预测状态。
由上可知，在预测节点状态的过程中，除了使用了从观测数据中获取的时间特征和静态空间特征，还使用了从观测数据中获取的动态空间特征，从而提高了具有动态拓扑关系的网络中的节点状态的预测准确度。
在训练过程中，可以通过最小化损失函数对时间模型、空间模型以及FC层进行联合训练，一种可选的损失函数如公式(8)所示。
Figure PCTCN2020141253-appb-000030
其中，
Figure PCTCN2020141253-appb-000031
为损失函数，x t+k为t+k时刻目标基站的真实状态，
Figure PCTCN2020141253-appb-000032
为均方误差(mean square error，MSE)，
Figure PCTCN2020141253-appb-000033
平均绝对误差(mean absolute error，MAE)，α用于平衡MSE和MAD的权重，N为节点的数量。
下面通过对比方法300的预测效果与其它方法的预测效果来说明本申请的有益效果。
参与对比的方法包括历史均值(historical average，HA)、预言模型(Prophet)、时空图卷积网络(spatio-temporal graph convolutional network，STGCN)、注意力时空图卷积网络(attention spatio-temporal graph convolutional network，ASTGCN)和扩散卷积循环神经网络(diffusion convolutional recurrent neural network，DCRNN)，评价指标为均方根误差(root mean square error，RMSE)和MAE，RMSE和MAE的值越小，说明方法的预测能力越好。对比结果如表1所示。
表1
Figure PCTCN2020141253-appb-000034
STGCN、ASTGCN、DCRNN和方法300都是基于深度学习的方法，每次运行的结果会有一些差异，表1给出了这几种方法分别预测10次的结果，并给出了均值和方差。由表1可以看出，方法300在几种方法中取得了最优的预测效果。
上文详细介绍了本申请提供的预测节点状态的方法的示例。可以理解的是，相应的装置为了实现上述功能，其包含了执行各个功能相应的硬件结构和/或软件模块。本领域技术人员应该很容易意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，本申请能够以硬件或硬件和计算机软件的结合形式来实现。某个功能究竟以硬件还是计算机软件驱动硬件的方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
本申请可以根据上述方法示例对装置进行功能单元的划分，例如，可以将各个功能划分为各个功能单元，也可以将两个或两个以上的功能集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。需要说明的是，本申 请中对单元的划分是示意性的，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式。
图7示出了本申请提供的一种预测节点状态的装置的结构示意图。装置700包括输入单元710和处理单元720。
处理单元720用于：获取目标网络中多个节点的静态图和动态图，所述静态图和所述动态图均为拓扑图；根据所述静态图和所述动态图，生成所述多个节点的空间特征数据；获取所述多个节点的时间特征数据；根据所述空间特征数据和所述时间特征数据，获取目标节点在目标时段内的预测状态，所述目标节点为所述多个节点中的任意一个节点。
可选地，输入单元710用于：获取所述多个节点在第一历史时段内的观测数据，所述第一历史时段为所述目标时段之前的时段；处理单元720具体用于：在时间维度上对所述观测数据进行切片，获取时间切片数据，所述时间切片数据的数据量小于所述观测数据的数据量；根据所述时间切片数据，获取所述时间特征数据。
可选地，所述时间切片数据包括：所述第一历史时段中与所述目标时段相邻的时段内的观测数据。
可选地，所述时间切片数据包括：所述第一历史时段中特定时段内的观测数据；其中，所述特定时段位于第一周期内，所述目标时段位于第二周期内，所述特定时段包括所述目标时段在所述第一周期内对应的时段。
可选地，输入单元710用于：获取所述多个节点在第二历史时段内的观测数据，所述第二历史时段为所述目标时段之前的时段；处理单元720具体用于：根据所述观测数据，构建所述静态图和所述动态图。
可选地，所述观测数据包括以下数据中的至少一个：气象数据、网络拓扑数据、流量数据、语音数据、信令数据、POI数据、重大事件数据、以及节假日数据。
可选地，处理单元720还用于：获取所述目标节点在所述目标时段内的真实状态；根据所述真实状态和所述预测状态训练时间模型和空间模型，其中，所述时间模型用于根据输入的观测数据输出所述时间特征数据，所述空间模型用于根据输入的所述静态图和所述动态图输出所述空间特征数据。
可选地，所述静态图属于第一类拓扑图，所述动态图属于第二类拓扑图，所述第一类拓扑图的拓扑关系变化速率小于所述第二类拓扑图的拓扑关系变化速率。
可选地，所述目标网络为通信网络，所述静态图为表征所述多个节点基于流量的拓扑关系的拓扑图，所述动态图为表征所述多个节点基于物理线路的拓扑关系的拓扑图。
装置700执行预测节点状态的方法的具体方式以及产生的有益效果可以参见方法实施例中的相关描述。
图8示出了本申请提供的一种预测节点状态的设备的结构示意图。图8中的虚线表示该单元或该模块为可选的。设备800可用于实现上述方法实施例中描述的方法。设备800可以是终端设备或服务器或芯片。
设备800包括一个或多个处理器801，该一个或多个处理器801可支持设备800实现方法实施例中的方法。处理器801可以是通用处理器或者专用处理器。例如，处理器801可以是中央处理器(central processing unit，CPU)。CPU可以用于对设备800进行控制，执行软件程序，处理软件程序的数据。设备800还可以包括通信单元805，用以实现信号 (如观测数据和预测结果)的输入(接收)和/或输出(发送)。
例如，设备800可以是芯片，通信单元805可以是该芯片的输入和/或输出电路，或者，通信单元805可以是该芯片的通信接口，该芯片可以作为终端设备或网络设备或其它电子设备的组成部分。
又例如，设备800可以是终端设备或服务器，通信单元805可以是该终端设备或该服务器的收发器，或者，通信单元805可以是该终端设备或该服务器的收发电路。
设备800中可以包括一个或多个存储器802，其上存有程序804，程序804可被处理器801运行，生成指令803，使得处理器801根据指令803执行上述方法实施例中描述的方法。可选地，存储器802中还可以存储有数据(如观测数据)。可选地，处理器801还可以读取存储器802中存储的数据，该数据可以与程序804存储在相同的存储地址，该数据也可以与程序804存储在不同的存储地址。
处理器801和存储器802可以单独设置，也可以集成在一起，例如，集成在终端设备的系统级芯片(system on chip，SOC)上。
处理器801执行方法实施例的具体方式可以参见方法实施例中的相关描述。
应理解，上述方法实施例的各步骤可以通过处理器801中的硬件形式的逻辑电路或者软件形式的指令完成。处理器801可以是CPU、数字信号处理器(digital signal processor，DSP)、专用集成电路(application specific integrated circuit，ASIC)、现场可编程门阵列(field programmable gate array，FPGA)或者其它可编程逻辑器件，例如，分立门、晶体管逻辑器件或分立硬件组件。
本申请还提供了一种计算机程序产品，该计算机程序产品被处理器801执行时实现本申请中任一方法实施例所述的方法。
该计算机程序产品可以存储在存储器802中，例如是程序804，程序804经过预处理、编译、汇编和链接等处理过程最终被转换为能够被处理器801执行的可执行目标文件。
本申请还提供了一种计算机可读存储介质，其上存储有计算机程序，该计算机程序被计算机执行时实现本申请中任一方法实施例所述的方法。该计算机程序可以是高级语言程序，也可以是可执行目标程序。
该计算机可读存储介质例如是存储器802。存储器802可以是易失性存储器或非易失性存储器，或者，存储器802可以同时包括易失性存储器和非易失性存储器。其中，非易失性存储器可以是只读存储器(read-only memory，ROM)、可编程只读存储器(programmable ROM，PROM)、可擦除可编程只读存储器(erasable PROM，EPROM)、电可擦除可编程只读存储器(electrically EPROM，EEPROM)或闪存。易失性存储器可以是随机存取存储器(random access memory，RAM)，其用作外部高速缓存。通过示例性但不是限制性说明，许多形式的RAM可用，例如静态随机存取存储器(static RAM，SRAM)、动态随机存取存储器(dynamic RAM，DRAM)、同步动态随机存取存储器(synchronous DRAM，SDRAM)、双倍数据速率同步动态随机存取存储器(double data rate SDRAM，DDR SDRAM)、增强型同步动态随机存取存储器(enhanced SDRAM，ESDRAM)、同步连接动态随机存取存储器(synchlink DRAM，SLDRAM)和直接内存总线随机存取存储器(direct rambus RAM，DR RAM)。
本领域的技术人员可以清楚地了解到，为了描述的方便和简洁，上述描述的装置和设 备的具体工作过程以及产生的技术效果，可以参考前述方法实施例中对应的过程和技术效果，在此不再赘述。
在本申请所提供的几个实施例中，所揭露的系统、装置和方法，可以通过其它的方式实现。例如，以上所描述的方法实施例的一些特征可以忽略，或不执行。以上所描述的装置实施例仅仅是示意性的，单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，多个单元或组件可以结合或者可以集成到另一个系统。另外，各单元之间的耦合或各个组件之间的耦合可以是直接耦合，也可以是间接耦合，上述耦合包括电的、机械的或其它形式的连接。
在本申请的各种实施例中，序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本申请的实施例的实施过程构成任何限定。
另外，本文中的术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本文中字符“/”，一般表示前后关联对象是一种“或”的关系。
总之，以上所述仅为本申请技术方案的较佳实施例而已，并非用于限定本申请的保护范围。凡在本申请的原则之内，所作的任何修改、等同替换、改进等，均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种预测节点状态的方法，其特征在于，包括：
    获取目标网络中多个节点的静态图和动态图，所述静态图和所述动态图均为拓扑图；
    根据所述静态图和所述动态图，生成所述多个节点的空间特征数据；
    获取所述多个节点的时间特征数据；
    根据所述空间特征数据和所述时间特征数据，获取目标节点在目标时段内的预测状态，所述目标节点为所述多个节点中的任意一个节点。
  2. 根据权利要求1所述的方法，其特征在于，所述获取所述多个节点的时间特征数据，包括：
    获取所述多个节点在第一历史时段内的观测数据，所述第一历史时段为所述目标时段之前的时段；
    在时间维度上对所述观测数据进行切片，获取时间切片数据，所述时间切片数据的数据量小于所述观测数据的数据量；
    根据所述时间切片数据，获取所述时间特征数据。
  3. 根据权利要求2所述的方法，其特征在于，所述时间切片数据包括：
    所述第一历史时段中与所述目标时段相邻的时段内的观测数据。
  4. 根据权利要求2或3所述的方法，其特征在于，所述时间切片数据包括：
    所述第一历史时段中特定时段内的观测数据，其中，所述特定时段位于第一周期内，所述目标时段位于第二周期内，所述特定时段包括所述目标时段在所述第一周期内对应的时段。
  5. 根据权利要求1至4中任一项所述的方法，其特征在于，所述获取多个节点的静态图和动态图，包括：
    获取所述多个节点在第二历史时段内的观测数据，所述第二历史时段为所述目标时段之前的时段；
    根据所述观测数据，构建所述静态图和所述动态图。
  6. 根据权利要求5所述的方法，其特征在于，所述观测数据包括以下数据中的至少一个：
    气象数据、网络拓扑数据、流量数据、语音数据、信令数据、兴趣点POI数据、重大事件数据、以及节假日数据。
  7. 根据权利要求1至6中任一项所述的方法，其特征在于，还包括：
    获取所述目标节点在所述目标时段内的真实状态；
    根据所述真实状态和所述预测状态训练时间模型和空间模型，其中，所述时间模型用于根据输入的观测数据输出所述时间特征数据，所述空间模型用于根据输入的所述静态图和所述动态图输出所述空间特征数据。
  8. 根据权利要求1至7中任一项所述的方法，其特征在于，所述静态图属于第一类拓扑图，所述动态图属于第二类拓扑图，所述第一类拓扑图的拓扑关系变化速率小于所述第二类拓扑图的拓扑关系变化速率。
  9. 根据权利要求1至8中任一项所述的方法，其特征在于，所述目标网络为通信网络，所述静态图为表征所述多个节点基于流量的拓扑关系的拓扑图，所述动态图为表征所述多个节点基于物理线路的拓扑关系的拓扑图。
  10. 一种预测节点状态的装置，其特征在于，包括处理单元，用于：
    获取目标网络中多个节点的静态图和动态图，所述静态图和所述动态图均为拓扑图；
    根据所述静态图和所述动态图，生成所述多个节点的空间特征数据；
    获取所述多个节点的时间特征数据；
    根据所述空间特征数据和所述时间特征数据，获取目标节点在目标时段内的预测状态，所述目标节点为所述多个节点中的任意一个节点。
  11. 根据权利要求10所述的装置，其特征在于，所述装置还包括输入单元，
    所述输入单元用于：获取所述多个节点在第一历史时段内的观测数据，所述第一历史时段为所述目标时段之前的时段；
    所述处理单元具体用于：在时间维度上对所述观测数据进行切片，获取时间切片数据，所述时间切片数据的数据量小于所述观测数据的数据量；根据所述时间切片数据，获取所述时间特征数据。
  12. 根据权利要求11所述的装置，其特征在于，所述时间切片数据包括：
    所述第一历史时段中与所述目标时段相邻的时段内的观测数据。
  13. 根据权利要求11或12所述的装置，其特征在于，所述时间切片数据包括：
    所述第一历史时段中特定时段内的观测数据，其中，所述特定时段位于第一周期内，所述目标时段位于第二周期内，所述特定时段包括所述目标时段在所述第一周期内对应的时段。
  14. 根据权利要求10至13中任一项所述的装置，其特征在于，所述装置还包括输入单元，
    所述输入单元用于：获取所述多个节点在第二历史时段内的观测数据，所述第二历史时段为所述目标时段之前的时段；
    所述处理单元具体用于：根据所述观测数据，构建所述静态图和所述动态图。
  15. 根据权利要求14所述的装置，其特征在于，所述观测数据包括以下数据中的至少一个：
    气象数据、网络拓扑数据、流量数据、语音数据、信令数据、兴趣点POI数据、重大事件数据、以及节假日数据。
  16. 根据权利要求10至15中任一项所述的装置，其特征在于，所述处理单元还用于：
    获取所述目标节点在所述目标时段内的真实状态；
    根据所述真实状态和所述预测状态训练时间模型和空间模型，其中，所述时间模型用于根据输入的观测数据输出所述时间特征数据，所述空间模型用于根据输入的所述静态图和所述动态图输出所述空间特征数据。
  17. 根据权利要求10至16中任一项所述的装置，其特征在于，所述静态图属于第一类拓扑图，所述动态图属于第二类拓扑图，所述第一类拓扑图的拓扑关系变化速率小于所述第二类拓扑图的拓扑关系变化速率。
  18. 根据权利要求10至17中任一项所述的装置，其特征在于，所述目标网络为通信 网络，所述静态图为表征所述多个节点基于流量的拓扑关系的拓扑图，所述动态图为表征所述多个节点基于物理线路的拓扑关系的拓扑图。
  19. 一种预测节点状态的设备，其特征在于，所述设备包括处理器和存储器，所述存储器用于存储计算机程序，所述处理器用于从所述存储器中调用并运行所述计算机程序，使得所述设备执行权利要求1至9中任一项所述的方法。
  20. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质存储了计算机程序，当所述计算机程序被处理器执行时，使得处理器执行权利要求1至9中任一项所述的方法。
PCT/CN2020/141253 2020-05-14 2020-12-30 预测节点状态的方法和装置 WO2021227535A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP20935951.2A EP4135264A4 (en) 2020-05-14 2020-12-30 METHOD AND DEVICE FOR PREDICTING A NODE STATUS
US17/984,421 US20230117633A1 (en) 2020-05-14 2022-11-10 Method and apparatus for predicting node state

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010409392.2 2020-05-14
CN202010409392.2A CN111726243B (zh) 2020-05-14 2020-05-14 预测节点状态的方法和装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/984,421 Continuation US20230117633A1 (en) 2020-05-14 2022-11-10 Method and apparatus for predicting node state

Publications (1)

Publication Number Publication Date
WO2021227535A1 true WO2021227535A1 (zh) 2021-11-18

Family

ID=72564499

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/141253 WO2021227535A1 (zh) 2020-05-14 2020-12-30 预测节点状态的方法和装置

Country Status (4)

Country Link
US (1) US20230117633A1 (zh)
EP (1) EP4135264A4 (zh)
CN (1) CN111726243B (zh)
WO (1) WO2021227535A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114417192A (zh) * 2022-03-28 2022-04-29 北京百度网讯科技有限公司 更新兴趣点poi状态的方法、装置、设备、介质及产品

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111726243B (zh) * 2020-05-14 2021-10-22 华为技术有限公司 预测节点状态的方法和装置
US20220343146A1 (en) * 2021-04-23 2022-10-27 Alibaba Singapore Holding Private Limited Method and system for temporal graph neural network acceleration
CN114827353B (zh) * 2022-04-15 2023-10-10 中国电信股份有限公司 通信网络通话预测方法、装置、设备及存储介质
CN114900441B (zh) * 2022-04-29 2024-04-26 华为技术有限公司 网络性能预测方法，性能预测模型训练方法及相关装置
CN115225546B (zh) * 2022-07-22 2023-11-28 北京天融信网络安全技术有限公司 一种网络流量的预测方法、装置及设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140133453A1 (en) * 2012-11-14 2014-05-15 Telefonaktiebolaget Lm Ericsson (Publ) Resource Allocation for Minimum Satisfaction Guarantees in Multi-Service and Multi-Antenna Networks
CN108234198A (zh) * 2017-12-19 2018-06-29 清华大学 一种基站流量预测方法和设备
CN109862585A (zh) * 2019-01-31 2019-06-07 湖北工业大学 一种基于深度时空神经网络的动态异构网络流量预测方法
CN110995520A (zh) * 2020-02-28 2020-04-10 清华大学 网络流量预测方法、装置、计算机设备及可读存储介质
CN111726243A (zh) * 2020-05-14 2020-09-29 华为技术有限公司 预测节点状态的方法和装置

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108053080B (zh) * 2017-12-30 2021-05-11 中国移动通信集团江苏有限公司 区域用户数量统计值预测方法、装置、设备及介质
CN108566305B (zh) * 2018-04-28 2021-04-06 中国人民大学 一种计算机网络智能组网与优化系统和方法
CN108879692B (zh) * 2018-06-26 2020-09-25 湘潭大学 一种区域综合能源系统能流分布预测方法及系统
US11221617B2 (en) * 2018-10-22 2022-01-11 Nec Corporation Graph-based predictive maintenance
CN110784903A (zh) * 2019-11-06 2020-02-11 周口师范学院 网络数据传输方法、装置、计算机设备和存储介质
CN110889546B (zh) * 2019-11-20 2020-08-18 浙江省交通规划设计研究院有限公司 一种基于注意力机制的交通流量模型训练方法
CN110969854A (zh) * 2019-12-13 2020-04-07 深圳先进技术研究院 一种交通流量的预测方法、系统及终端设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140133453A1 (en) * 2012-11-14 2014-05-15 Telefonaktiebolaget Lm Ericsson (Publ) Resource Allocation for Minimum Satisfaction Guarantees in Multi-Service and Multi-Antenna Networks
CN108234198A (zh) * 2017-12-19 2018-06-29 清华大学 一种基站流量预测方法和设备
CN109862585A (zh) * 2019-01-31 2019-06-07 湖北工业大学 一种基于深度时空神经网络的动态异构网络流量预测方法
CN110995520A (zh) * 2020-02-28 2020-04-10 清华大学 网络流量预测方法、装置、计算机设备及可读存储介质
CN111726243A (zh) * 2020-05-14 2020-09-29 华为技术有限公司 预测节点状态的方法和装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114417192A (zh) * 2022-03-28 2022-04-29 北京百度网讯科技有限公司 更新兴趣点poi状态的方法、装置、设备、介质及产品

Also Published As

Publication number Publication date
CN111726243B (zh) 2021-10-22
US20230117633A1 (en) 2023-04-20
CN111726243A (zh) 2020-09-29
EP4135264A4 (en) 2023-09-13
EP4135264A1 (en) 2023-02-15

Similar Documents

Publication Publication Date Title
WO2021227535A1 (zh) 预测节点状态的方法和装置
CN111932036B (zh) 基于位置大数据的精细时空尺度动态人口预测方法及系统
CN110969854A (zh) 一种交通流量的预测方法、系统及终端设备
WO2023103587A1 (zh) 短临降水预测方法及装置
US20220092418A1 (en) Training method for air quality prediction model, prediction method and apparatus, device, program, and medium
US20130304363A1 (en) Identifying purpose-based origin-destination using call detailed records
CN106935034A (zh) 面向车联网的区域交通流量预测系统及方法
CN113610286B (zh) 顾及时空相关性和气象因素的pm2.5浓度预测方法及装置
US20240160923A1 (en) Increasing Accuracy and Resolution of Weather Forecasts Using Deep Generative Models
CN116106988A (zh) 天气预测方法、装置、电子设备及存储介质
CN113496310A (zh) 基于深度学习模型的大气污染物预测方法和系统
Wang et al. Adaptive multi-receptive field spatial-temporal graph convolutional network for traffic forecasting
Gao et al. A deep learning framework with spatial-temporal attention mechanism for cellular traffic prediction
CN116090504A (zh) 图神经网络模型训练方法及装置、分类方法、计算设备
CN114418243B (zh) 分布式新能源云端网格预测方法与系统
CN115310732B (zh) 航班延误预测方法及系统
Yalçın Weather parameters forecasting with time series using deep hybrid neural networks
Zhong et al. Probabilistic fine-grained urban flow inference with normalizing flows
Lin et al. Integrating ANFIS and Qt Framework to Develop a Mobile-Based Typhoon Rainfall Forecasting System
CN114548572A (zh) 城市路网交通状态的预测方法、装置、设备及介质
CN114566048A (zh) 一种基于多视角自适应时空图网络的交通控制方法
CN113836302A (zh) 文本分类方法、文本分类装置及存储介质
CN113255352A (zh) 一种街道信息确定方法、装置及计算机设备
CN114726463A (zh) 基于神经网络的移动通信用户时空分布预测方法及装置
Hu et al. A simplified deep residual network for citywide crowd flows prediction

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: 20935951

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020935951

Country of ref document: EP

Effective date: 20221110

NENP Non-entry into the national phase

Ref country code: DE