WO2022088677A1 - 建立区域热度预测模型、区域热度预测的方法及装置 - Google Patents

建立区域热度预测模型、区域热度预测的方法及装置 Download PDF

Info

Publication number
WO2022088677A1
WO2022088677A1 PCT/CN2021/097892 CN2021097892W WO2022088677A1 WO 2022088677 A1 WO2022088677 A1 WO 2022088677A1 CN 2021097892 W CN2021097892 W CN 2021097892W WO 2022088677 A1 WO2022088677 A1 WO 2022088677A1
Authority
WO
WIPO (PCT)
Prior art keywords
time
heat
prediction model
historical
data
Prior art date
Application number
PCT/CN2021/097892
Other languages
English (en)
French (fr)
Inventor
黄际洲
王海峰
范淼
孙一博
Original Assignee
北京百度网讯科技有限公司
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 北京百度网讯科技有限公司 filed Critical 北京百度网讯科技有限公司
Priority to US17/622,950 priority Critical patent/US20220414691A1/en
Priority to EP21827170.8A priority patent/EP4012581A4/en
Priority to JP2021576943A priority patent/JP2023502817A/ja
Priority to KR1020217042754A priority patent/KR20220058858A/ko
Publication of WO2022088677A1 publication Critical patent/WO2022088677A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/906—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/08—Thermal analysis or thermal optimisation

Definitions

  • the present application relates to the field of computer application technology, and in particular, to a method and apparatus for establishing a regional heat prediction model and regional heat prediction in the field of big data technology.
  • Regional heat prediction is of great value, allowing governments, institutions and individuals to more effectively optimize resource allocation, predict development trends, and even provide reference or advice on transportation.
  • the so-called regional heat prediction refers to predicting the flow of people in an area at a designated time, such as predicting the passenger flow of a business district at a designated time, predicting the passenger flow of a station at a designated time, and so on.
  • Regional heat prediction is essentially a time series prediction problem.
  • Existing time series prediction algorithms include feature engineering, neural network fitting and other algorithms.
  • the existing time series prediction algorithms are highly dependent on a large amount of labeled data, and can only achieve high prediction accuracy after training on data with a long stationary sequence history.
  • some small-probability events will have an impact on the regional heat, which will cause the regional heat to fluctuate sharply over a period of time compared to usual.
  • due to the lack of historical data for such small-probability events traditional methods cannot effectively learn them, resulting in poor regional heat prediction accuracy in the event of small-probability events.
  • the present application aims to provide a method and apparatus for establishing a regional heat prediction model and regional heat prediction, so as to solve the above-mentioned technical problems.
  • the present application provides a method for establishing a regional heat prediction model, including:
  • the second support sample is used to further train the time series prediction model, so as to adjust the model parameters to obtain the regional heat prediction model;
  • the regional heat prediction model is used to predict the second query sample, where the second query sample includes the regional heat at the time to be predicted.
  • the present application provides a method for regional heat prediction, including:
  • the regional heat prediction model uses the first historical regional heat data to pre-train the time series prediction model, uses the second historical regional heat data as the second support sample, and uses the second support sample to perform the time series prediction model on the time series prediction model. obtained from further training.
  • the present application provides a device for establishing a regional heat prediction model, including:
  • the pre-training module is used to pre-train the time series prediction model by using the first historical area heat data
  • a fine-tuning module configured to use the second historical regional heat data as a second support sample, and use the second support sample to further train the time series prediction model, so as to adjust the model parameters to obtain the regional heat prediction model;
  • the regional heat prediction model is used to predict the second query sample, where the second query sample includes the regional heat at the time to be predicted.
  • the present application provides an apparatus for regional heat prediction, including:
  • a prediction module configured to use a regional heat prediction model to predict a second query sample, where the second query sample includes the regional heat at the time to be predicted;
  • the regional heat prediction model uses the first historical regional heat data to pre-train the time series prediction model, uses the second historical regional heat data as the second support sample, and uses the second support sample to perform the time series prediction model on the time series prediction model. obtained from further training.
  • the application provides an electronic device, comprising:
  • the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.
  • the present application provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the above method.
  • the present application uses the second historical area heat data as a support sample to fine-tune the time series prediction model, so as to establish a model capable of 2.
  • the regional heat prediction model for predicting the regional heat of the to-be-predicted time in the query sample. This method enables effective learning of the distribution data even when the distribution data represented by the second historical area heat data is small, thereby improving the accuracy of the area heat prediction.
  • Embodiment 1 is a flow chart of a main method provided by Embodiment 1 of the present application;
  • FIG. 2 is a schematic structural diagram of a time series prediction model provided in Embodiment 1 of the present application.
  • FIG. 3 is a data example diagram of a business district heat prediction provided by Embodiment 1 of the present application.
  • FIG. 4 is a data example diagram of a business district heat prediction provided in Embodiment 2 of the present application.
  • FIG. 5 is a structural diagram of an apparatus for establishing a regional heat prediction model provided by an embodiment of the present application.
  • FIG. 6 is a structural diagram of an apparatus for regional heat prediction provided by an embodiment of the present application.
  • FIG. 7 is a block diagram of an electronic device used to implement an embodiment of the present application.
  • the learning technology based on small samples can use a large amount of conventionally distributed historical data to train a time series prediction model based on small samples, and then train the trained model.
  • a small amount of irregularly distributed historical data satisfactory prediction accuracy can be achieved on a small amount of irregularly distributed data, which is more ideal than traditional supervised learning algorithms.
  • the data of Few-shot Learning consists of support samples (Support Set) and query samples (Query Set).
  • the Support Set contains instances of known categories, and the Query Set contains more than one unlabeled category of data.
  • the task of Few-shot Learning is to predict the data categories in the Query Set.
  • FIG. 1 the flow chart of the main method provided in Embodiment 1 of the present application is shown in FIG. 1 , and may include the following steps:
  • a time series prediction model is pre-trained by using the first historical area heat data.
  • the second historical area heat data is used as a second support sample, and the time series prediction model is further trained by using the second support sample, so as to adjust the model parameters to obtain the area heat prediction model.
  • the first historical region heat data may include feature data and region heatness of the region at each time point in the first historical period.
  • the second historical area heat data may include feature data and area heat at each time point of the area in the second historical period.
  • the area involved in the embodiment of the present application may be an AOI (Area of Interest, a surface of interest), a ROI (Regin of Interest, an area of interest) in a geographic location, or may be a plurality of POIs (Point Of Interest, a point of interest) ), AOI, ROI, etc.
  • AOI Absolute of Interest
  • ROI Area of interest
  • POIs Point Of Interest, a point of interest
  • the characteristic data may include at least one of the number of POIs containing points of interest in the area, user visit time distribution data, visiting user travel mode distribution data, and visiting user travel distance distribution data.
  • Regional popularity largely reflects the number of users who visit the region.
  • step 103 can be performed by using the regional heat prediction model obtained by training, that is, the second query sample is predicted by using the regional heat prediction model, and the second query sample includes the regional heat at the time to be predicted. .
  • the above-mentioned first historical area heat data may be commonly distributed historical area heat data.
  • the second historical area heat data may be uncommonly distributed historical area heat data.
  • the prediction task is the passenger flow at ⁇ time points in the future
  • x t can be obtained from characteristic data such as the number of POIs included in the business district, user visit time distribution data, visiting user travel mode distribution data, and visiting user travel distance distribution data.
  • Each feature is converted into a one-hot (one-hot) vector, and these vectors are concatenated to form a feature vector x t with n dimensions, where n is a positive integer.
  • a business district consists of some POIs of commercial places, then the number of POIs contained in the business district can form a one-dimensional vector.
  • the visiting time distribution of the business district can be represented by the number of visiting users in 24 hours a day, which constitutes a 24-dimensional vector.
  • the travel mode distribution data of visiting users in this business district can be represented by the number of people who travel by various means of transportation. Assuming that the means of transportation are public transportation, bicycles, private cars and walking, this feature can be represented as a 4 dimensional vector.
  • the travel distance distribution data of the visiting users in the business circle can be represented by the number of different travel distances from the departure to the destination business circle. If the travel distance is discretized into 10 grades, namely: ⁇ 0.25km, 0.25km-0.5km, 0.5km-1km, 1km-2km, 2km-5km, 5km-6km, 6km-10km, 10km-15km, 15km-30km . >30km, then the vector can be represented as a 10-dimensional vector.
  • a 39-dimensional feature vector x t can be obtained.
  • the feature vector obtained from the above features can fully reflect the customer flow situation reflected by the business district at a specific time point, so that the subsequent prediction of the business district's popularity is more accurate.
  • the time series prediction model adopted in this embodiment may be based on a recurrent neural network.
  • the model structure can be as shown in FIG. 2 .
  • the feature at each time point is mapped to an m-dimensional hidden layer state vector h using a recurrent neural network t , which can be expressed as:
  • F is a nonlinear function, which can be LSTM (Long Short-Term Memory, long short-term memory network) or GRU (Gated Recurrent Unit, gated recurrent unit) and other recurrent neural networks with the ability to model sequences.
  • LSTM Long Short-Term Memory, long short-term memory network
  • GRU Gated Recurrent Unit
  • the hidden layer state vector h T at the last moment in the historical period T is sent to the feedforward neural network to obtain the prediction output Expressed as:
  • W and b are model parameters, where W is a vector of m ⁇ dimension, and b is a vector of ⁇ dimension.
  • time series prediction model When training the above-mentioned time series prediction model, firstly, pre-training is performed by using the feature data and the popularity of the business district at each time point in the first historical period T1 of the business district. Then, the time series prediction model obtained by the pre-training is further trained by using the characteristic data of the business district at each time point in the second historical period T2 and the popularity of the business district, so as to adjust the model parameters.
  • the training in the process of pre-training by using the feature data and business district popularity at each time point in the first historical period T1 of the business district, the training can be constructed using the characteristic data and business district popularity at each time point in the first historical period T1 sample.
  • a time window Tw may be used.
  • the length of Tw is shorter than the duration of T1 and less than the duration of T2.
  • the length of the time window Tw is consistent with the duration of the historical period T in the above prediction task. Usually, it is much smaller than the duration of T1 and the duration of T2, for example, the duration of 5 time points is selected as the time window Tw .
  • a training sample is constructed using the feature data and the business district heat at each time point in the time window Tw and the business district heat at ⁇ time points after the time window Tw .
  • N training samples can be generated, where N is a positive integer.
  • the time series prediction model uses the feature data and business district heat at each time point in the time window Tw to predict the business district heat at ⁇ time points after the time window Tw .
  • the training goal is to minimize the prediction results and training
  • the loss function can be constructed by using the above training objective, that is, by using the difference between the prediction result and the hotness of the business district at ⁇ time points after the time window Tw in the training sample.
  • the mean square error loss function can be constructed by using the prediction results and the heat of the business circle at ⁇ time points after the time window Tw in the training sample, and the value of the loss function can be used for back-propagation to optimize the parameters of the time series prediction model until it reaches Training end condition.
  • the value of the loss function is less than or equal to a preset loss function threshold, or the number of iterations reaches a preset threshold, and so on.
  • the loss function L can take the following formula:
  • N is the number of training samples
  • Y i is the business district heat at ⁇ time points after the time window Tw in the i- th training sample value.
  • the above-mentioned time window is also used.
  • the model parameters are further optimized and adjusted on the basis of the model parameters of the time series prediction model obtained by pre-training.
  • the time series prediction model after the above pre-training and adjustment can be used to predict the popularity of the business district at some time points in the second query sample.
  • the part of the time point refers to the time to be predicted, and the duration of the time to be predicted is less than or equal to the above ⁇ .
  • the heat data of each week in a certain business district in 2018 and 2019 can be used as the first historical business district heat data, that is, by using the weekly heat data of a certain business district in 2018 and 2019.
  • the heat data is used as a training task to construct training samples and pre-train the time series prediction model. Since a well-known small probability event occurred in 2020, the heat data from 1 to 8 weeks in 2020 can be used as the heat data of the second historical business district, and the time series prediction model can be further trained to adjust the model parameters. After adjustment, the regional heat prediction model is obtained to predict the regional heat at the time to be predicted in the 9th to 26th week of 2020 (ie, the second query sample).
  • This embodiment is further improved on the basis of the second embodiment, and uses the meta-learning method to pre-train the time series prediction model, so as to obtain model parameters that can be quickly fitted on the second support sample.
  • At least one meta-training task is constructed by using the first historical area heat data.
  • Each meta-training task includes the first support sample and the first query sample, and the meta-learning mechanism is used to train the time series prediction model.
  • a meta-training task is constructed using the business district popularity data in 2018, and another meta-training task is constructed using the business district popularity data in 2019.
  • support samples and query samples are divided in each training task.
  • the support samples and query samples in the pre-training task are referred to as "first support samples” and "first query samples”
  • the support samples in the subsequent fine-tuning tasks are called “second support samples”
  • the query samples in the prediction task are called “second query samples”.
  • “first” and “second” do not have the meaning of quantity and order, but are only used to distinguish the names.
  • the training task actually corresponds to the meta-training (meta-train) process in the meta-learning mechanism
  • the fine-tuning task and the prediction task actually correspond to the meta-testing (meta-test) process in the meta-learning mechanism
  • the first support sample and the first query sample have the same duration as the second support sample and the second query sample, respectively.
  • the business district heat data from the 1st to 8th week in mid-2018 is used as the first support sample in the first meta-training task
  • the business district heat data from the 9th to 26th week is used as the first support sample in the first meta-training task.
  • Query samples Take the business district heat data from the 1st to the 8th week in 2019 as the first support sample in the second meta-training task, and the business district heat data from the 9th to the 26th week as the first query sample.
  • the difference from the second embodiment is that in the meta-train process, for each meta-learning task, the gradient is calculated on the first support sample and the meta-parameters are updated, and the loss function of all meta-learning tasks on the first query sample is calculated.
  • the gradient is used to update the model parameters until the end-of-training condition is reached. For example, the value of the loss function is less than or equal to a preset loss function threshold, or the number of iterations reaches a preset threshold, and so on.
  • the pre-training is completed, and the pre-trained time series prediction model is obtained.
  • the second support sample for example, the business district heat data from the end of 2020 in Figure 4
  • the gradient of the loss function on the second support sample is calculated and used to update the model parameters.
  • the adjustment process is similar to the subsequent prediction process of the region heat at the time to be predicted in the second query sample, which is similar to that in the second embodiment, and will not be repeated.
  • the training samples are constructed in a similar manner as in the second embodiment, that is, the feature data and the business district heat at each time point within the time window Tw and the business district heat at ⁇ time points after the time window Tw are used.
  • Build a training sample A plurality of corresponding training samples can be generated by sliding the time window Tw among the corresponding first support samples, first query samples, and second support samples.
  • the length of the time window Tw is shorter than the duration of the first support sample, and is shorter than the duration of the first query sample.
  • the prediction results can be displayed, stored in a specific format, and further analyzed.
  • the prediction result can also be sent to the user terminal actively or in response to a request of the user terminal.
  • the apparatus is an application that can be located on the server side, or can also be a plug-in or a software development kit (Software Development Kit) in the application located on the server side. , SDK) and other functional units, or may also be located in a computer terminal, which is not particularly limited in this embodiment of the present invention.
  • the apparatus may include: a pre-training module 01 and a fine-tuning module 02 .
  • the main functions of each unit are as follows:
  • the pre-training module 01 is used for pre-training the time series prediction model by using the first historical area heat data.
  • the fine-tuning module 02 is configured to use the second historical regional heat data as a second support sample, and use the second support sample to further train the time series prediction model, so as to adjust the model parameters to obtain the regional heat prediction model.
  • the regional heat prediction model is used to predict the second query sample, and the second query sample includes the regional heat at the time to be predicted.
  • the first historical area heat data includes feature data and area heat at each time point of the area in the first historical period
  • the second historical area heat data includes feature data and area heat at each time point of the area in the second historical period.
  • the feature data includes at least one of the number of POIs that include points of interest in the area, user visit time distribution data, visiting user travel mode distribution data, and visiting user travel distance distribution data.
  • the training objectives of the time series prediction model include: minimizing the difference between the prediction result and the expected value, and the prediction result is that the time series model uses the feature data of each time point in the time window Tw and the regional heat to ⁇ after the time window Tw.
  • the result of the prediction of the regional heat at the time point, the expected value is the regional heat in the corresponding sample at ⁇ time points after the time window Tw ; the time window Tw is less than the length of the first historical period and less than the length of the second historical period, ⁇ is a positive integer.
  • the pre-training module 01 can use the first historical area heat data to construct at least one meta-training task; based on the first support sample and the first query sample in the meta-training task, the meta-learning mechanism is used to train the time sequence A prediction model; wherein the first support sample and the first query sample have the same duration as the second support sample and the second query sample respectively.
  • the pre-training module 01 adopts the meta-learning mechanism to train the time-series prediction model, for each meta-learning task, it can calculate the gradient of the loss function on the first support sample and update the meta-parameters; determine that all meta-learning tasks are in the first query sample The gradient of the loss function on and used to update the model parameters;
  • the fine-tuning module 02 is specifically used to calculate the gradient of the loss function on the second support sample and to update the model parameters.
  • the loss function can be used by the time series model to use the feature data and regional heat of each time point in the time window Tw to predict the regional heat of ⁇ time points after the time window Tw , and then use the prediction result and the time window Tw after ⁇ The difference between the regional heats in the corresponding samples at each time point is constructed;
  • the length of the time window Tw is shorter than the duration of the first support sample and shorter than the duration of the first query sample, and ⁇ is a positive integer.
  • FIG. 6 is a structural diagram of a device for regional heat prediction provided by an embodiment of the present application
  • the device is an application that can be located on the server side, or can also be a plug-in or a software development kit (Software Development Kit, SDK) in the application located on the server side. ) and other functional units, or may also be located in a computer terminal, which is not particularly limited in this embodiment of the present invention.
  • the apparatus may include:
  • the prediction module 11 is configured to use the regional heat prediction model to predict the second query sample, where the second query sample includes the regional heat at the time to be predicted.
  • the regional heat prediction model is obtained by using the first historical regional heat data to pre-train the time series prediction model, using the second historical regional heat data as the second support sample, and using the second support sample to further train the time series prediction model. That is, it is obtained by pre-training using the device shown in FIG. 5 .
  • the first historical area heat data includes the characteristic data and area heat at each time point of the area in the first historical period
  • the second historical area heat data includes feature data and area heat at each time point of the area in the second historical period.
  • the prediction module 11 can use the feature data and regional heat of each time point in the time window T before the time to be predicted to predict the regional heat of the time to be predicted; the time window T is less than the length of the first historical period , and is less than the length of the second historical period.
  • the above-mentioned time window T is actually the same length as the time window Tw used in the process of training the time series prediction model in the fourth embodiment, and the total duration of the time to be predicted needs to be less than or equal to the duration of ⁇ used in the process of training the time series prediction model.
  • the present application further provides an electronic device and a readable storage medium.
  • FIG. 7 it is a block diagram of an electronic device for establishing a regional heat prediction model or a method for regional heat prediction according to an embodiment of the present application.
  • Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers.
  • Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices.
  • the components shown herein, their connections and relationships, and their functions are by way of example only, and are not intended to limit implementations of the application described and/or claimed herein.
  • the electronic device includes: one or more processors 701 , a memory 702 , and interfaces for connecting various components, including a high-speed interface and a low-speed interface.
  • the various components are interconnected using different buses and may be mounted on a common motherboard or otherwise as desired.
  • the processor may process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of the GUI on an external input/output device, such as a display device coupled to the interface.
  • multiple processors and/or multiple buses may be used with multiple memories and multiple memories, if desired.
  • multiple electronic devices may be connected, with each device providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system).
  • a processor 701 is taken as an example in FIG. 7 .
  • the memory 702 is the non-transitory computer-readable storage medium provided by the present application.
  • the memory stores instructions executable by at least one processor, so that the at least one processor executes the method for establishing an area heat prediction model or a method for area heat prediction provided by the present application.
  • the non-transitory computer-readable storage medium of the present application stores computer instructions for causing a computer to execute the method for establishing an area heat prediction model or a method for area heat prediction provided by the present application.
  • the memory 702 can be used to store non-transitory software programs, non-transitory computer-executable programs and modules, such as the method for establishing a regional heat prediction model or a method for regional heat prediction in the embodiments of the present application.
  • Corresponding program instruction/module The processor 701 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions and modules stored in the memory 702, that is, to realize the establishment of the regional heat prediction model or the regional heat prediction in the above method embodiments. method method.
  • the memory 702 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the electronic device, and the like. Additionally, memory 702 may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory 702 may optionally include memory located remotely from processor 701, which may be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
  • the electronic device may further include: an input device 703 and an output device 704 .
  • the processor 701 , the memory 702 , the input device 703 and the output device 704 may be connected by a bus or in other ways, and the connection by a bus is taken as an example in FIG. 7 .
  • the input device 703 can receive input numerical or character information, and generate key signal input related to user settings and function control of the electronic device, such as a touch screen, keypad, mouse, trackpad, touchpad, pointing stick, one or more Input devices such as mouse buttons, trackballs, joysticks, etc.
  • Output devices 704 may include display devices, auxiliary lighting devices (eg, LEDs), haptic feedback devices (eg, vibration motors), and the like.
  • the display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some implementations, the display device may be a touch screen.
  • Various implementations of the systems and techniques described herein can be implemented in digital electronic circuitry, integrated circuit systems, application specific ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpretable on a programmable system including at least one programmable processor that The processor, which may be a special purpose or general-purpose programmable processor, may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device an output device.
  • the processor which may be a special purpose or general-purpose programmable processor, may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device an output device.
  • machine-readable medium and “computer-readable medium” refer to any computer program product, apparatus, and/or apparatus for providing machine instructions and/or data to a programmable processor ( For example, magnetic disks, optical disks, memories, programmable logic devices (PLDs), including machine-readable media that receive machine instructions as machine-readable signals.
  • machine-readable signal refers to any signal used to provide machine instructions and/or data to a programmable processor.
  • the systems and techniques described herein may be implemented on a computer having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user ); and a keyboard and pointing device (eg, a mouse or trackball) through which a user can provide input to the computer.
  • a display device eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor
  • a keyboard and pointing device eg, a mouse or trackball
  • Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (eg, visual feedback, auditory feedback, or tactile feedback); and can be in any form (including acoustic input, voice input, or tactile input) to receive input from the user.
  • the systems and techniques described herein may be implemented on a computing system that includes back-end components (eg, as a data server), or a computing system that includes middleware components (eg, an application server), or a computing system that includes front-end components (eg, a user's computer having a graphical user interface or web browser through which a user may interact with implementations of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system.
  • the components of the system may be interconnected by any form or medium of digital data communication (eg, a communication network). Examples of communication networks include: Local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.
  • a computer system can include clients and servers.
  • Clients and servers are generally remote from each other and usually interact through a communication network.
  • the relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Databases & Information Systems (AREA)
  • Strategic Management (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Human Resources & Organizations (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • General Business, Economics & Management (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Operations Research (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Probability & Statistics with Applications (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请公开了一种建立区域热度预测模型、区域热度预测的方法及装置，涉及大数据技术领域。具体实现方案为：利用第一历史区域热度数据，预训练时序预测模型；将第二历史区域热度数据作为第二支持样本，利用所述第二支持样本对所述时序预测模型进行进一步训练，以对模型参数进行调整得到所述区域热度预测模型；所述区域热度预测模型用于对第二查询样本进行预测，所述第二查询样本包括待预测时间的区域热度。该方式使得在第二历史区域热度数据所代表的分布数据较少的情况下，也能够对该分布数据进行有效学习，从而提高区域热度预测的准确度。

Description

建立区域热度预测模型、区域热度预测的方法及装置
本申请要求了申请日为2020年10月26日，申请号为2020111550161发明名称为“建立区域热度预测模型、区域热度预测的方法及装置”的中国专利申请的优先权。
技术领域
本申请涉及计算机应用技术领域，特别涉及大数据技术领域下一种建立区域热度预测模型、区域热度预测的方法及装置。
背景技术
区域热度预测具有巨大的价值，可以让政府、机构和个人更加有效地优化资源配置，预判发展趋势，甚至在交通出行方面提供参考或建议。所谓区域热度预测指的是预测区域在指定时间的人流量，例如预测某个商圈在指定时间的客流量，预测某个车站在指定时间的客流量，等等。
区域热度预测本质上是一个时间序列预测问题，现有的时间序列预测算法包括特征工程、神经网络拟合等算法。然而，现有的时间序列预测算法对大量标注数据存在高度依赖，只能在具有较长的平稳性序列历史记录的数据上进行训练后，才能够具有较高的预测准确率。然而一些小概率事件发生时会对区域热度产生影响，从而导致在一段时间内区域热度相较于平时发生剧烈波动。然而，由于这类小概率事件的历史数据较少，采用传统方法则无法对此进行有效学习，从而导致发生小概率事件情况下的区域热度预测准确度很差。
发明内容
有鉴于此，本申请旨在提供一种建立区域热度预测模型、区域热度预测的方法及装置，以解决上述技术问题。
第一方面，本申请提供了一种建立区域热度预测模型的方法，包括：
利用第一历史区域热度数据，预训练时序预测模型；
将第二历史区域热度数据作为第二支持样本，利用所述第二支持样本对所述时序预测模型进行进一步训练，以对模型参数进行调整得到所 述区域热度预测模型；
所述区域热度预测模型用于对第二查询样本进行预测，所述第二查询样本包括待预测时间的区域热度。
第二方面，本申请提供了一种区域热度预测的方法，包括：
利用区域热度预测模型对第二查询样本进行预测，所述第二查询样本包括待预测时间的区域热度；
其中所述区域热度预测模型是利用第一历史区域热度数据预训练时序预测模型后，将第二历史区域热度数据作为第二支持样本，并利用所述第二支持样本对所述时序预测模型进行进一步训练得到的。
第三方面，本申请提供了一种建立区域热度预测模型的装置，包括：
预训练模块，用于利用第一历史区域热度数据，预训练时序预测模型；
微调模块，用于将第二历史区域热度数据作为第二支持样本，利用所述第二支持样本对所述时序预测模型进行进一步训练，以对模型参数进行调整得到所述区域热度预测模型；
所述区域热度预测模型用于对第二查询样本进行预测，所述第二查询样本包括待预测时间的区域热度。
第四方面，本申请提供了一种区域热度预测的装置，包括：
预测模块，用于利用区域热度预测模型对第二查询样本进行预测，所述第二查询样本包括待预测时间的区域热度；
其中所述区域热度预测模型是利用第一历史区域热度数据预训练时序预测模型后，将第二历史区域热度数据作为第二支持样本，并利用所述第二支持样本对所述时序预测模型进行进一步训练得到的。
第五方面，本申请提供了一种电子设备，包括：
至少一个处理器；以及
与所述至少一个处理器通信连接的存储器；其中，
所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行上述的方法。
第六方面，本申请提供了一种存储有计算机指令的非瞬时计算机可读存储介质，其中，所述计算机指令用于使所述计算机执行上述的方法。
由以上技术方案可以看出，本申请在利用第一历史区域热度数据预训练得到的时序预测模型基础上，将第二历史区域热度数据作为支持样本对时序预测模型进行微调，从而建立能够对第二查询样本中待预测时间的区域热度进行预测的区域热度预测模型。该方式使得在第二历史区域热度数据所代表的分布数据较少的情况下，也能够对该分布数据进行有效学习，从而提高区域热度预测的准确度。
上述可选方式所具有的其他效果将在下文中结合具体实施例加以说明。
附图说明
附图用于更好地理解本方案，不构成对本申请的限定。其中：
图1为本申请实施例一提供的主要方法流程图；
图2是本申请实施例一提供的一种时序预测模型的结构示意图；
图3为本申请实施例一提供的一种商圈热度预测的数据实例图；
图4为本申请实施例二提供的一种商圈热度预测的数据实例图；
图5为本申请实施例提供的一种建立区域热度预测模型的装置结构图；
图6为本申请实施例提供的一种区域热度预测的装置结构图；
图7是用来实现本申请实施例的电子设备的框图。
具体实施方式
以下结合附图对本申请的示范性实施例做出说明，其中包括本申请实施例的各种细节以助于理解，应当将它们认为仅仅是示范性的。因此，本领域普通技术人员应当认识到，可以对这里描述的实施例做出各种改变和修改，而不会背离本申请的范围和精神。同样，为了清楚和简明，以下的描述中省略了对公知功能和结构的描述。
本申请实施例中采用了基于小样本的学习技术(Few-shot Learning)，基于小样本的学习技术可以利用大量常规分布的历史数据训练基于小样本的时间序列预测模型，再将训练好的模型配合少量不常规分布的历史数据，就能在少量不常规分布的数据上取得令人满意的预测准确率，比传统监督学习算法具有更加理想的效果。
Few-shot Learning的数据由支持样本(Support Set)和查询样本(Query Set)组成。Support Set中包含已知类别的实例，Query Set中包含一条以上未标注类别的数据，Few-shot Learning的任务就是对Query Set中的数据类别进行预测。
实施例一
基于上述理论，本申请实施例一提供的主要方法流程图如图1中所示，可以包括以下步骤：
在101中，利用第一历史区域热度数据，预训练时序预测模型。
在102中，将第二历史区域热度数据作为第二支持样本，利用第二支持样本对时序预测模型进行进一步训练，以对模型参数进行调整得到区域热度预测模型。
在本申请实施例中，第一历史区域热度数据可以包括第一历史时段中区域在各时间点的特征数据和区域热度。第二历史区域热度数据可以包括第二历史时段中区域在各时间点的特征数据和区域热度。
其中，本申请实施例所涉及的区域可以是地理位置上的AOI(Area of Interest，兴趣面)、ROI(Regin of Interest，兴趣区域)，也可以是由多个POI(Point Of Interest，兴趣点)、AOI、ROI等形成的区域。例如可以是商圈、校区、科技园区，等等。
其中特征数据可以包括区域包含兴趣点POI的数量、用户到访时间分布数据、到访用户的出行方式分布数据以及到访用户的出行距离分布数据中的至少一种。区域热度很大程度上体现的是到访该区域的用户量。
更进一步地，训练得到区域热度预测模型后，可以利用训练得到的区域热度预测模型执行步骤103，即利用区域热度预测模型对第二查询样本进行预测，第二查询样本包括待预测时间的区域热度。
作为其中一种实施方式，上述第一历史区域热度数据可以是常见分布的历史区域热度数据。第二历史区域热度数据可以是不常见分布的历史区域热度数据。
下面以商圈热度预测为例，结合两个实施例对本申请提供的上述方法进行详细描述。
实施例二
假设目前的预测任务为：已知历史时段T中特定商圈各时间点的区 域热度数据，其中该商圈在该历史时段T中的特征可以用一个特征序列X＝(x 1,x 2,…,x T)表示，其中，x t表示该商圈在时间点t的特征向量。相应地，商圈在历史时段T中的客流量即热度，可以用Y＝(y 1,y 2,…,y T)表示，y t表示该商圈在时间点t的热度。对于已知X和Y后，预测任务是未来τ个时间点的客流量
Figure PCTCN2021097892-appb-000001
其中，x t可以由商圈包含POI的数量、用户到访时间分布数据、到访用户的出行方式分布数据以及到访用户的出行距离分布数据等特征数据得到。每一种特征都被转换成一个one-hot(独热)向量，这些向量拼接后构成一个具有n个维度的特征向量x t，n为正整数。
举个例子：
某商圈由一些商业场所类的POI构成，那么该商圈包含POI的数量可以构成一个一维向量。
该商圈的到访时间分布可以采用每天24个小时的到访用户数量来表示，即构成一个24维的向量。
该商圈到访用户的出行方式分布数据可以采用到访人群中使用各种交通工具出行的数量来表示，假设交通工具采用公共交通、自行车、私家车和步行，那么该特征可以表示成一个4维的向量。
该商圈到访用户的出行距离分布数据可以采用从出发达到目的地商圈的不同出行距离的数量来表示。如果将出行距离离散化为10档，即：<0.25km,0.25km-0.5km,0.5km-1km,1km-2km,2km-5km,5km-6km,6km-10km,10km-15km,15km-30km。>30km，那么该向量可以表示成一个10维的向量。
然后将各向量进行拼接后，可以得到一个39维的特征向量x t。由上述特征得到的特征向量能够充分反映商圈在特定时间点所反映出的客流量状况，使得后续对于商圈热度的预测更加准确。
作为其中一种实现方式，本实施例中采用的时序预测模型可以基于循环神经网络。模型结构可以如图2中所示。
对于预测任务而言，对于给定输入的特征序列X＝(x 1,x 2,…,x T)，使用循环神经网络将每一个时间点的特征映射为一个m维度的隐层状态向量h t，可以表示为：
h t＝F(x t,h t-1)      (1)
其中，F是一个非线性的函数，可以是LSTM(Long Short-Term Memory，长短期记忆网络)或者GRU(Gated Recurrent Unit，门控循环单元)等具有建模序列能力的循环神经网络。
将历史时段T中最后一个时刻的隐层状态向量h T送入前馈神经网络，得到预测输出
Figure PCTCN2021097892-appb-000002
表示为：
Figure PCTCN2021097892-appb-000003
其中，W和b为模型参数，其中，W为m×τ维的向量，b为τ维的向量。
在进行上述时序预测模型的训练时，首先利用该商圈第一历史时段T1中各时间点的特征数据和商圈热度进行预训练。然后利用该商圈在第二历史时段T2中各时间点的特征数据和商圈热度对预训练得到的时序预测模型进行进一步训练，以对模型参数进行调整。
其中，在利用该商圈第一历史时段T1中各时间点的特征数据和商圈热度进行预训练的过程中，可以利用第一历史时段T1中各时间点的特征数据和商圈热度构建训练样本。可以采用一个时间窗T w，T w的长度小于T1的时长，且小于T2的时长，该时间窗T w的长度与上述预测任务中历史时段T的时长一致。通常远小于T1的时长和T2的时长，例如选取5个时间点的时长作为时间窗T w。在构建训练样本时，利用时间窗T w内各时间点的特征数据和商圈热度以及该时间窗T w之后τ个时间点的商圈热度构建一个训练样本。将时间窗T w在第一历史时段T1中进行滑动就可以产生N训练样本，N为正整数。
训练过程中，时序预测模型利用时间窗T w内各时间点的特征数据和商圈热度对时间窗T w之后τ个时间点的商圈热度进行预测，训练目标为：最小化预测结果与训练样本中时间窗T w之后τ个时间点的商圈热度之间的差值。损失函数可以利用上述训练目标进行构建，即利用预测结果与训练样本中时间窗T w之后τ个时间点的商圈热度之间的差值来构建。例如可以利用预测结果和训练样本中时间窗T w之后τ个时间点的商圈热度构建均方误差损失函数，利用损失函数的值进行反向传播以对时序预测模型的参数进行优化，直至达到训练结束条件。例如，损失函数的值小于或等于预设损失函数阈值，或者，迭代次数达到预设次数阈值，等等。
例如，损失函数L可以采用如下公式：
Figure PCTCN2021097892-appb-000004
其中，N为训练样本的数量，
Figure PCTCN2021097892-appb-000005
为时序预测模型对第i个训练样本中时间窗T w之后τ个时间点的商圈热度的预测值，Y i为第i个训练样本中时间窗T w之后τ个时间点的商圈热度值。
预训练时序预测模型之后，在利用第二历史时段T2中各时间点的特征数据和商圈热度(即第二支持样本)对预训练得到的时序预测模型进行进一步训练时，同样采用上述时间窗的方式构建训练样本，不同的是，在预训练得到的时序预测模型的模型参数基础上，对模型参数进行进一步优化调整。
经过上述预训练和调整之后的时序预测模型，可以用于对第二查询样本中部分时间点的商圈热度的预测。其中，所述部分时间点指的是待预测时间，该待预测时间的时长小于或等于上述τ。在进行预测时，将待预测时间之前时间段T中各时间点的特征数据和商圈热度输入时序预测模型，就可以得到时序预测模型输出的待预测时间的区域热度。预测方式即采用上述公式(1)和(2)。
举个例子，如图3中所示，可以将2018年、2019年中某商圈各周的热度数据作为第一历史商圈热度数据，即利用2018年、2019年中某商圈各周的热度数据作为一个训练任务构建训练样本，对时序预测模型进行预训练。由于2020年发生了众所周知的小概率事件，因此，可以将2020年1～8周的热度数据作为第二历史商圈热度数据，对时序预测模型进行进一步训练，以调整模型参数。调整后得到区域热度预测模型用以预测2020年第9～26周(即第二查询样本)中待预测时间的区域热度。
实施例三
本实施例在实施例二的基础上进行进一步改进，使用了元学习的方法对时序预测模型进行预训练，从而得到可以在第二支持样本上进行迅速拟合的模型参数。
在本实施例中，在预训练之前，利用第一历史区域热度数据构建至少一个元训练任务。每个元训练任务中都包含第一支持样本和第一查询样本，采用元学习机制训练时序预测模型。
举个例子，如图4中所示，利用2018年的商圈热度数据构建一个元训练任务，将2019年的商圈热度数据构建另一个元训练任务。其中，每 个训练任务中都进行了支持样本和查询样本的划分。为了与微调任务和预测任务中的支持样本和查询样本进行区分，在本申请实施例中，将预训练任务中的支持样本和查询样本称为“第一支持样本”和“第一查询样本”，将后续微调任务中的支持样本称为“第二支持样本”，预测任务中的查询样本称为“第二查询样本”。其中“第一”和“第二”并不具备数量和顺序上的含义，仅仅用于在名称上进行区分。另外，本实施例中，与训练任务实际上对应了元学习机制中的元训练(meta-train)过程，微调任务和预测任务实际上对应了元学习机制中的元测试(meta-test)过程。
其中第一支持样本和第一查询样本分别与第二支持样本、第二查询样本的时长一致。如图4中所示，将2018年中第1～8周的商圈热度数据作为第一个元训练任务中的第一支持样本，将第9～26周的商圈热度数据在作为第一查询样本。将2019年中第1～8周的商圈热度数据作为第二个元训练任务中的第一支持样本，将第9～26周的商圈热度数据在作为第一查询样本。
与实施例二中不同的是，在meta-train过程中，对于每个元学习任务，在第一支持样本上计算梯度并更新元参数，计算所有元学习任务在第一查询样本上的损失函数梯度并用以更新模型参数，直至达到训练结束条件。例如，损失函数的值小于或等于预设损失函数阈值，或者，迭代次数达到预设次数阈值，等等。至此完成预训练，得到预训练后的时序预测模型。
然后在预训练得到的时序预测模型上利用第二支持样本(例如图4中2020年底1～8周的商圈热度数据)对时序预测模型进行进一步训练，调整模型参数。在此过程中计算在第二支持样本上的损失函数梯度并用以更新模型参数。该调整过程与后续对第二查询样本中待预测时间的区域热度的预测过程与实施例二中类似，不做赘述。
在上述训练过程中采用与实施例二中类似的方式构建训练样本，即利用时间窗T w内各时间点的特征数据和商圈热度以及该时间窗T w之后τ个时间点的商圈热度构建一个训练样本。将时间窗T w在对应的第一支持样本、第一查询样本、第二支持样本中进行滑动就可以产生对应的多个训练样本。在本实施例中，时间窗T w的长度小于第一支持样本的时长， 且小于第一查询样本的时长。训练目标以及构建的损失函数参见实施例二中的相关记载，在此不做赘述。
通过上述实施例中的方式，可以在发生小概率事件时，帮助模型很好地理解小概率事件对区域所造成的持续性影响，更加准确地预测区域热度，从而帮助用户更加有针对性地优化资源配置，预判发展趋势。
在实现上述对待预测时间的区域热度预测后，可以将预测结果进行显示、特定格式的存储、进一步分析等处理。还可以主动或者响应于用户终端的请求，将预测结果发送给用户终端。
以上是对本申请所提供方法进行的详细描述，下面结合实施例对本申请提供的装置进行详细描述。
实施例四
图5为本申请实施例提供的一种建立区域热度预测模型的装置结构图，该装置是可以位于服务器端的应用，或者还可以为位于服务器端的应用中的插件或软件开发工具包(Software Development Kit，SDK)等功能单元，或者，还可以位于计算机终端，本发明实施例对此不进行特别限定。如图5中所示，该装置可以包括：预训练模块01和微调模块02。其中各组成单元的主要功能如下：
预训练模块01，用于利用第一历史区域热度数据，预训练时序预测模型。
微调模块02，用于将第二历史区域热度数据作为第二支持样本，利用第二支持样本对时序预测模型进行进一步训练，以对模型参数进行调整得到区域热度预测模型。
区域热度预测模型用于对第二查询样本进行预测，第二查询样本包括待预测时间的区域热度。
作为一种优选的实施方式，第一历史区域热度数据包括第一历史时段中区域在各时间点的特征数据和区域热度；
第二历史区域热度数据包括第二历史时段中区域在各时间点的特征数据和区域热度。
其中，特征数据包括区域包含兴趣点POI的数量、用户到访时间分布数据、到访用户的出行方式分布数据以及到访用户的出行距离分布数据中至少一种。
其中，时序预测模型的训练目标包括：最小化预测结果与期望值之间的差值，预测结果为时序模型利用时间窗T w内各时间点的特征数据和区域热度对时间窗T w之后τ个时间点的区域热度进行预测的结果，期望值为时间窗T w之后τ个时间点在对应样本中的区域热度；时间窗T w小于第一历史时段的长度，且小于第二历史时段的长度，τ为正整数。
作为一种优选的实施方式，预训练模块01可以利用第一历史区域热度数据，构建至少一个元训练任务；基于元训练任务中的第一支持样本和第一查询样本，采用元学习机制训练时序预测模型；其中第一支持样本、第一查询样本分别与第二支持样本、第二查询样本的时长一致。
其中，预训练模块01在采用元学习机制训练时序预测模型时，可以对于各元学习任务，在第一支持样本上计算损失函数的梯度并更新元参数；确定所有元学习任务在第一查询样本上的损失函数梯度并用以更新模型参数；
微调模块02，具体用于在第二支持样本上的计算损失函数梯度并用以更新模型参数。
其中，损失函数可以由时序模型利用时间窗T w内各时间点的特征数据和区域热度对时间窗T w之后τ个时间点的区域热度进行预测后，利用预测结果与时间窗T w之后τ个时间点在对应样本中的区域热度之间的差值构建得到；
其中，时间窗T w的长度小于第一支持样本的时长、且小于第一查询样本的时长，τ为正整数。
实施例五
图6为本申请实施例提供的一种区域热度预测的装置结构图，该装置是可以位于服务器端的应用，或者还可以为位于服务器端的应用中的插件或软件开发工具包(Software Development Kit，SDK)等功能单元，或者，还可以位于计算机终端，本发明实施例对此不进行特别限定。如图6中所示，该装置可以包括：
预测模块11，用于利用区域热度预测模型对第二查询样本进行预测，第二查询样本包括待预测时间的区域热度。
其中区域热度预测模型是利用第一历史区域热度数据预训练时序预测模型后，将第二历史区域热度数据作为第二支持样本，并利用第二支 持样本对时序预测模型进行进一步训练得到的。即采用如图5中所示的装置预先训练得到。
其中，第一历史区域热度数据包括第一历史时段中区域在各时间点的特征数据和区域热度；
第二历史区域热度数据包括第二历史时段中区域在各时间点的特征数据和区域热度。
作为一种优选的实施方式，预测模块11可以利用待预测时间之前时间窗T内各时间点的特征数据和区域热度，对待预测时间的区域热度进行预测；时间窗T小于第一历史时段的长度，且小于第二历史时段的长度。
上述时间窗T实际上与实施例四中训练时序预测模型过程中所采用的时间窗T w长度一致，待预测时间的总时长需要小于或等于训练时序预测模型过程中所采用的τ的时长。
根据本申请的实施例，本申请还提供了一种电子设备和一种可读存储介质。
如图7所示，是根据本申请实施例的建立区域热度预测模型或区域热度预测的方法的电子设备的框图。电子设备旨在表示各种形式的数字计算机，诸如，膝上型计算机、台式计算机、工作台、个人数字助理、服务器、刀片式服务器、大型计算机、和其它适合的计算机。电子设备还可以表示各种形式的移动装置，诸如，个人数字处理、蜂窝电话、智能电话、可穿戴设备和其它类似的计算装置。本文所示的部件、它们的连接和关系、以及它们的功能仅仅作为示例，并且不意在限制本文中描述的和/或者要求的本申请的实现。
如图7所示，该电子设备包括：一个或多个处理器701、存储器702，以及用于连接各部件的接口，包括高速接口和低速接口。各个部件利用不同的总线互相连接，并且可以被安装在公共主板上或者根据需要以其它方式安装。处理器可以对在电子设备内执行的指令进行处理，包括存储在存储器中或者存储器上以在外部输入/输出装置(诸如，耦合至接口的显示设备)上显示GUI的图形信息的指令。在其它实施方式中，若需要，可以将多个处理器和/或多条总线与多个存储器和多个存储器一起使用。同样，可以连接多个电子设备，各个设备提供部分必要的操作(例 如，作为服务器阵列、一组刀片式服务器、或者多处理器系统)。图7中以一个处理器701为例。
存储器702即为本申请所提供的非瞬时计算机可读存储介质。其中，所述存储器存储有可由至少一个处理器执行的指令，以使所述至少一个处理器执行本申请所提供的建立区域热度预测模型或区域热度预测的方法的方法。本申请的非瞬时计算机可读存储介质存储计算机指令，该计算机指令用于使计算机执行本申请所提供的建立区域热度预测模型或区域热度预测的方法的方法。
存储器702作为一种非瞬时计算机可读存储介质，可用于存储非瞬时软件程序、非瞬时计算机可执行程序以及模块，如本申请实施例中的建立区域热度预测模型或区域热度预测的方法的方法对应的程序指令/模块。处理器701通过运行存储在存储器702中的非瞬时软件程序、指令以及模块，从而执行服务器的各种功能应用以及数据处理，即实现上述方法实施例中的建立区域热度预测模型或区域热度预测的方法的方法。
存储器702可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储根据该电子设备的使用所创建的数据等。此外，存储器702可以包括高速随机存取存储器，还可以包括非瞬时存储器，例如至少一个磁盘存储器件、闪存器件、或其他非瞬时固态存储器件。在一些实施例中，存储器702可选包括相对于处理器701远程设置的存储器，这些远程存储器可以通过网络连接至该电子设备。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
该电子设备还可以包括：输入装置703和输出装置704。处理器701、存储器702、输入装置703和输出装置704可以通过总线或者其他方式连接，图7中以通过总线连接为例。
输入装置703可接收输入的数字或字符信息，以及产生与该电子设备的用户设置以及功能控制有关的键信号输入，例如触摸屏、小键盘、鼠标、轨迹板、触摸板、指示杆、一个或者多个鼠标按钮、轨迹球、操纵杆等输入装置。输出装置704可以包括显示设备、辅助照明装置(例如，LED)和触觉反馈装置(例如，振动电机)等。该显示设备可以包 括但不限于，液晶显示器(LCD)、发光二极管(LED)显示器和等离子体显示器。在一些实施方式中，显示设备可以是触摸屏。
此处描述的系统和技术的各种实施方式可以在数字电子电路系统、集成电路系统、专用ASIC(专用集成电路)、计算机硬件、固件、软件、和/或它们的组合中实现。这些各种实施方式可以包括：实施在一个或者多个计算机程序中，该一个或者多个计算机程序可在包括至少一个可编程处理器的可编程系统上执行和/或解释，该可编程处理器可以是专用或者通用可编程处理器，可以从存储系统、至少一个输入装置、和至少一个输出装置接收数据和指令，并且将数据和指令传输至该存储系统、该至少一个输入装置、和该至少一个输出装置。
这些计算程序(也称作程序、软件、软件应用、或者代码)包括可编程处理器的机器指令，并且可以利用高级过程和/或面向对象的编程语言、和/或汇编/机器语言来实施这些计算程序。如本文使用的，术语“机器可读介质”和“计算机可读介质”指的是用于将机器指令和/或数据提供给可编程处理器的任何计算机程序产品、设备、和/或装置(例如，磁盘、光盘、存储器、可编程逻辑装置(PLD))，包括，接收作为机器可读信号的机器指令的机器可读介质。术语“机器可读信号”指的是用于将机器指令和/或数据提供给可编程处理器的任何信号。
为了提供与用户的交互，可以在计算机上实施此处描述的系统和技术，该计算机具有：用于向用户显示信息的显示装置(例如，CRT(阴极射线管)或者LCD(液晶显示器)监视器)；以及键盘和指向装置(例如，鼠标或者轨迹球)，用户可以通过该键盘和该指向装置来将输入提供给计算机。其它种类的装置还可以用于提供与用户的交互；例如，提供给用户的反馈可以是任何形式的传感反馈(例如，视觉反馈、听觉反馈、或者触觉反馈)；并且可以用任何形式(包括声输入、语音输入或者、触觉输入)来接收来自用户的输入。
可以将此处描述的系统和技术实施在包括后台部件的计算系统(例如，作为数据服务器)、或者包括中间件部件的计算系统(例如，应用服务器)、或者包括前端部件的计算系统(例如，具有图形用户界面或者网络浏览器的用户计算机，用户可以通过该图形用户界面或者该网络浏览器来与此处描述的系统和技术的实施方式交互)、或者包括这种后 台部件、中间件部件、或者前端部件的任何组合的计算系统中。可以通过任何形式或者介质的数字数据通信(例如，通信网络)来将系统的部件相互连接。通信网络的示例包括：局域网(LAN)、广域网(WAN)和互联网。
计算机系统可以包括客户端和服务器。客户端和服务器一般远离彼此并且通常通过通信网络进行交互。通过在相应的计算机上运行并且彼此具有客户端-服务器关系的计算机程序来产生客户端和服务器的关系。
应该理解，可以使用上面所示的各种形式的流程，重新排序、增加或删除步骤。例如，本发申请中记载的各步骤可以并行地执行也可以顺序地执行也可以不同的次序执行，只要能够实现本申请公开的技术方案所期望的结果，本文在此不进行限制。
上述具体实施方式，并不构成对本申请保护范围的限制。本领域技术人员应该明白的是，根据设计要求和其他因素，可以进行各种修改、组合、子组合和替代。任何在本申请的精神和原则之内所作的修改、等同替换和改进等，均应包含在本申请保护范围之内。

Claims (22)

  1. 一种建立区域热度预测模型的方法，包括：
    利用第一历史区域热度数据，预训练时序预测模型；
    将第二历史区域热度数据作为第二支持样本，利用所述第二支持样本对所述时序预测模型进行进一步训练，以对模型参数进行调整得到所述区域热度预测模型；
    所述区域热度预测模型用于对第二查询样本进行预测，所述第二查询样本包括待预测时间的区域热度。
  2. 根据权利要求1所述的方法，其中，所述第一历史区域热度数据包括第一历史时段中区域在各时间点的特征数据和区域热度；
    所述第二历史区域热度数据包括第二历史时段中所述区域在各时间点的特征数据和区域热度。
  3. 根据权利要求2所述的方法，其中，所述特征数据包括以下至少一种：
    所述区域包含兴趣点POI的数量、用户到访时间分布数据、到访用户的出行方式分布数据以及到访用户的出行距离分布数据。
  4. 根据权利要求2所述的方法，其中，所述时序预测模型的训练目标包括：最小化预测结果与期望值之间的差值，所述预测结果为所述时序模型利用时间窗T w内各时间点的特征数据和区域热度对时间窗T w之后τ个时间点的区域热度进行预测的结果，所述期望值为时间窗T w之后τ个时间点在对应样本中的区域热度；
    所述时间窗T w小于所述第一历史时段的长度，且小于所述第二历史时段的长度，所述τ为正整数。
  5. 根据权利要求1所述的方法，其中，所述利用第一历史区域热度数据，预训练时序预测模型包括：
    利用所述第一历史区域热度数据，构建至少一个元训练任务；
    基于元训练任务中的第一支持样本和第一查询样本，采用元学习机制训练所述时序预测模型；
    其中所述第一支持样本、第一查询样本分别与所述第二支持样本、第二查询样本的时长一致。
  6. 根据权利要求5所述的方法，其中，所述采用元学习机制训练所述时序预测模型包括：对于各元学习任务，在所述第一支持样本上计算损失函数的梯度并更新元参数；确定所有元学习任务在第一查询样本上的损失函数梯度并用以更新模型参数；
    利用所述第二支持样本对所述时序预测模型进行进一步训练，以对模型参数进行调整包括：在所述第二支持样本上的计算损失函数梯度并用以更新模型参数。
  7. 根据权利要求6所述的方法，其中，所述损失函数是由所述时序模型利用时间窗T w内各时间点的特征数据和区域热度对时间窗T w之后τ个时间点的区域热度进行预测后，利用预测结果与时间窗T w之后τ个时间点在对应样本中的区域热度之间的差值构建得到；
    其中，所述时间窗T w的长度小于所述第一支持样本的时长、且小于所述第一查询样本的时长，所述τ为正整数。
  8. 一种区域热度预测的方法，包括：
    利用区域热度预测模型对第二查询样本进行预测，所述第二查询样本包括待预测时间的区域热度；
    其中所述区域热度预测模型是利用第一历史区域热度数据预训练时序预测模型后，将第二历史区域热度数据作为第二支持样本，并利用所述第二支持样本对所述时序预测模型进行进一步训练得到的。
  9. 根据权利要求8所述的方法，其中，所述第一历史区域热度数据包括第一历史时段中区域在各时间点的特征数据和区域热度；
    所述第二历史区域热度数据包括第二历史时段中所述区域在各时间点的特征数据和区域热度。
  10. 根据权利要求9所述的方法，其中，所述利用区域热度预测模型对第二查询样本进行预测包括：
    利用所述待预测时间之前时间窗T内各时间点的特征数据和区域热度，对所述待预测时间的区域热度进行预测；
    所述时间窗T小于所述第一历史时段的长度，且小于所述第二历史时段的长度。
  11. 一种建立区域热度预测模型的装置，包括：
    预训练模块，用于利用第一历史区域热度数据，预训练时序预测模 型；
    微调模块，用于将第二历史区域热度数据作为第二支持样本，利用所述第二支持样本对所述时序预测模型进行进一步训练，以对模型参数进行调整得到所述区域热度预测模型；
    所述区域热度预测模型用于对第二查询样本进行预测，所述第二查询样本包括待预测时间的区域热度。
  12. 根据权利要求11所述的装置，其中，所述第一历史区域热度数据包括第一历史时段中区域在各时间点的特征数据和区域热度；
    所述第二历史区域热度数据包括第二历史时段中所述区域在各时间点的特征数据和区域热度。
  13. 根据权利要求12所述的装置，其中，所述特征数据包括以下至少一种：
    所述区域包含兴趣点POI的数量、用户到访时间分布数据、到访用户的出行方式分布数据以及到访用户的出行距离分布数据。
  14. 根据权利要求12所述的装置，其中，所述时序预测模型的训练目标包括：最小化预测结果与期望值之间的差值，所述预测结果为所述时序模型利用时间窗T w内各时间点的特征数据和区域热度对时间窗T w之后τ个时间点的区域热度进行预测的结果，所述期望值为时间窗T w之后τ个时间点在对应样本中的区域热度；
    所述时间窗T w小于所述第一历史时段的长度，且小于所述第二历史时段的长度，所述τ为正整数。
  15. 根据权利要求11所述的装置，其中，所述预训练模块，具体用于：
    利用所述第一历史区域热度数据，构建至少一个元训练任务；
    基于元训练任务中的第一支持样本和第一查询样本，采用元学习机制训练所述时序预测模型；
    其中所述第一支持样本、第一查询样本分别与所述第二支持样本、第二查询样本的时长一致。
  16. 根据权利要求15所述的装置，其中，所述预训练模块在采用元学习机制训练所述时序预测模型时，具体用于：对于各元学习任务，在所述第一支持样本上计算损失函数的梯度并更新元参数；确定所有元学 习任务在第一查询样本上的损失函数梯度并用以更新模型参数；
    所述微调模块，具体用于在所述第二支持样本上的计算损失函数梯度并用以更新模型参数。
  17. 根据权利要求16所述的装置，其中，所述损失函数是由所述时序模型利用时间窗T w内各时间点的特征数据和区域热度对时间窗T w之后τ个时间点的区域热度进行预测后，利用预测结果与时间窗T w之后τ个时间点在对应样本中的区域热度之间的差值构建得到；
    其中，所述时间窗T w的长度小于所述第一支持样本的时长、且小于所述第一查询样本的时长，所述τ为正整数。
  18. 一种区域热度预测的装置，包括：
    预测模块，用于利用区域热度预测模型对第二查询样本进行预测，所述第二查询样本包括待预测时间的区域热度；
    其中所述区域热度预测模型是利用第一历史区域热度数据预训练时序预测模型后，将第二历史区域热度数据作为第二支持样本，并利用所述第二支持样本对所述时序预测模型进行进一步训练得到的。
  19. 根据权利要求18所述的装置，其中，所述第一历史区域热度数据包括第一历史时段中区域在各时间点的特征数据和区域热度；
    所述第二历史区域热度数据包括第二历史时段中所述区域在各时间点的特征数据和区域热度。
  20. 根据权利要求19所述的装置，其中，所述预测模块，具体用于利用所述待预测时间之前时间窗T内各时间点的特征数据和区域热度，对所述待预测时间的区域热度进行预测；
    所述时间窗T小于所述第一历史时段的长度，且小于所述第二历史时段的长度。
  21. 一种电子设备，包括：
    至少一个处理器；以及
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行权利要求1-10中任一项所述的方法。
  22. 一种存储有计算机指令的非瞬时计算机可读存储介质，其中， 所述计算机指令用于使所述计算机执行权利要求1-10中任一项所述的方法。
PCT/CN2021/097892 2020-10-26 2021-06-02 建立区域热度预测模型、区域热度预测的方法及装置 WO2022088677A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
US17/622,950 US20220414691A1 (en) 2020-10-26 2021-06-02 Methods and apparatuses for regional heat prediction model establishment and regional heat prediction
EP21827170.8A EP4012581A4 (en) 2020-10-26 2021-06-02 METHODS AND DEVICES FOR CREATING A REGIONAL HEAT FORECAST MODEL AND REGIONAL HEAT FORECAST
JP2021576943A JP2023502817A (ja) 2020-10-26 2021-06-02 領域熱度予測モデルを確立する方法、領域熱度予測の方法、及び装置
KR1020217042754A KR20220058858A (ko) 2020-10-26 2021-06-02 구역 인기 예측 모델을 구축하는 방법, 구역 인기 예측 방법 및 장치

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011155016.1 2020-10-26
CN202011155016.1A CN112269930B (zh) 2020-10-26 2020-10-26 建立区域热度预测模型、区域热度预测的方法及装置

Publications (1)

Publication Number Publication Date
WO2022088677A1 true WO2022088677A1 (zh) 2022-05-05

Family

ID=74341385

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/097892 WO2022088677A1 (zh) 2020-10-26 2021-06-02 建立区域热度预测模型、区域热度预测的方法及装置

Country Status (6)

Country Link
US (1) US20220414691A1 (zh)
EP (1) EP4012581A4 (zh)
JP (1) JP2023502817A (zh)
KR (1) KR20220058858A (zh)
CN (1) CN112269930B (zh)
WO (1) WO2022088677A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117422960A (zh) * 2023-12-14 2024-01-19 广州华微明天软件技术有限公司 一种基于元学习的图像识别持续学习方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112269930B (zh) * 2020-10-26 2023-10-24 北京百度网讯科技有限公司 建立区域热度预测模型、区域热度预测的方法及装置
CN113343082A (zh) * 2021-05-25 2021-09-03 北京字节跳动网络技术有限公司 可热字段预测模型生成方法、装置、存储介质及设备
CN113643066A (zh) * 2021-08-16 2021-11-12 京东城市(北京)数字科技有限公司 客流量推断模型的训练方法以及推断客流量的方法和装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108345857A (zh) * 2018-02-09 2018-07-31 北京天元创新科技有限公司 一种基于深度学习的区域人群密度预测方法及装置
JP2019040475A (ja) * 2017-08-25 2019-03-14 日本電信電話株式会社 人流予測装置、システムおよびプログラム
CN110826695A (zh) * 2019-10-30 2020-02-21 京东数字城市(成都)科技有限公司 数据处理方法、装置和计算机可读存储介质
CN110852447A (zh) * 2019-11-15 2020-02-28 腾讯云计算（北京）有限责任公司 元学习方法和装置、初始化方法、计算设备和存储介质
CN111144648A (zh) * 2019-12-25 2020-05-12 中国联合网络通信集团有限公司 人流量预测设备及方法
CN112269930A (zh) * 2020-10-26 2021-01-26 北京百度网讯科技有限公司 建立区域热度预测模型、区域热度预测的方法及装置

Family Cites Families (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2013211616A (ja) * 2012-03-30 2013-10-10 Sony Corp 端末装置、端末制御方法、プログラム、および情報処理システム
JP6708122B2 (ja) * 2014-06-30 2020-06-10 日本電気株式会社 誘導処理装置及び誘導方法
CN105206040B (zh) * 2015-08-07 2017-06-23 北京航空航天大学 一种基于ic卡数据的公交串车预测方法
CN106952105A (zh) * 2017-04-26 2017-07-14 浙江大学 一种基于迁移学习的商铺最优化选址方法
JP6802118B2 (ja) * 2017-07-04 2020-12-16 株式会社日立製作所 情報処理システム
CN108306962B (zh) * 2018-01-30 2021-09-28 河海大学常州校区 一种商业大数据分析系统
CN108564228A (zh) * 2018-04-26 2018-09-21 重庆大学 一种基于时序特征预测轨道交通od客流量的方法
US20200125955A1 (en) * 2018-10-23 2020-04-23 International Business Machines Corporation Efficiently learning from highly-diverse data sets
WO2020158217A1 (ja) * 2019-02-01 2020-08-06 ソニー株式会社 情報処理装置、情報処理方法及び情報処理プログラム
CN111652364A (zh) * 2019-03-04 2020-09-11 富士通株式会社 训练元学习网络的装置和方法
CN110188668B (zh) * 2019-05-28 2020-09-25 复旦大学 一种面向小样本视频动作分类的方法
CN110348601A (zh) * 2019-06-06 2019-10-18 华南理工大学 一种基于双向长短期记忆网络的地铁短期客流量预测方法
CN110569886B (zh) * 2019-08-20 2023-02-28 天津大学 一种双向通道注意力元学习的图像分类方法
CN111047085B (zh) * 2019-12-06 2022-09-06 北京理工大学 一种基于元学习的混合动力车辆工况预测方法
CN111667095A (zh) * 2020-04-30 2020-09-15 百度在线网络技术（北京）有限公司 预测经济状态、建立经济状态预测模型的方法及对应装置
CN111401558B (zh) * 2020-06-05 2020-10-09 腾讯科技（深圳）有限公司 数据处理模型训练方法、数据处理方法、装置、电子设备
CN111680837B (zh) * 2020-06-08 2023-12-08 北京化工大学 一种竞争环境下的多场景多商品连锁便利店选址优化方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2019040475A (ja) * 2017-08-25 2019-03-14 日本電信電話株式会社 人流予測装置、システムおよびプログラム
CN108345857A (zh) * 2018-02-09 2018-07-31 北京天元创新科技有限公司 一种基于深度学习的区域人群密度预测方法及装置
CN110826695A (zh) * 2019-10-30 2020-02-21 京东数字城市(成都)科技有限公司 数据处理方法、装置和计算机可读存储介质
CN110852447A (zh) * 2019-11-15 2020-02-28 腾讯云计算（北京）有限责任公司 元学习方法和装置、初始化方法、计算设备和存储介质
CN111144648A (zh) * 2019-12-25 2020-05-12 中国联合网络通信集团有限公司 人流量预测设备及方法
CN112269930A (zh) * 2020-10-26 2021-01-26 北京百度网讯科技有限公司 建立区域热度预测模型、区域热度预测的方法及装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117422960A (zh) * 2023-12-14 2024-01-19 广州华微明天软件技术有限公司 一种基于元学习的图像识别持续学习方法
CN117422960B (zh) * 2023-12-14 2024-03-26 广州华微明天软件技术有限公司 一种基于元学习的图像识别持续学习方法

Also Published As

Publication number Publication date
KR20220058858A (ko) 2022-05-10
US20220414691A1 (en) 2022-12-29
CN112269930B (zh) 2023-10-24
CN112269930A (zh) 2021-01-26
JP2023502817A (ja) 2023-01-26
EP4012581A4 (en) 2022-08-24
EP4012581A1 (en) 2022-06-15

Similar Documents

Publication Publication Date Title
WO2022088677A1 (zh) 建立区域热度预测模型、区域热度预测的方法及装置
JP7166322B2 (ja) モデルを訓練するための方法、装置、電子機器、記憶媒体およびコンピュータプログラム
KR20220003085A (ko) 검색 결과를 결정하는 방법, 장치, 기기 및 컴퓨터 기록 매체
US20220019341A1 (en) Map information display method and apparatus, electronic device, and computer storage medium
WO2021232724A1 (zh) 提取地理位置点空间关系的方法、训练提取模型的方法及装置
CN113094550A (zh) 视频检索方法、装置、设备和介质
US11593384B2 (en) Parking lot free parking space predicting method, apparatus, electronic device and storage medium
US11574259B2 (en) Parking lot free parking space predicting method, apparatus, electronic device and storage medium
CN111708876A (zh) 生成信息的方法和装置
US11829447B2 (en) Resident area prediction method, apparatus, device, and storage medium
KR102630243B1 (ko) 구두점 예측 방법 및 장치
CN111931067A (zh) 兴趣点推荐方法、装置、设备和介质
WO2023184777A1 (zh) 更新兴趣点poi状态的方法、装置、设备、介质及产品
JP2021174560A (ja) 経済状態の予測方法、経済状態予測モデルの構築方法及び対応装置
CN113657934A (zh) 客流量预测模型的训练方法和客流量的预测方法和装置
CN113642740A (zh) 模型训练方法及装置、电子设备和介质
CN111611808A (zh) 用于生成自然语言模型的方法和装置
JP2022006166A (ja) 地図上の目的地の決定方法、機器、及び記憶媒体
CN111582452B (zh) 生成神经网络模型的方法和装置
CN111461306B (zh) 特征评估的方法及装置
CN111047107B (zh) 公路通行时间预测方法、装置、电子设备和存储介质
CN112819497B (zh) 转化率预测方法、装置、设备和存储介质
CN112598136A (zh) 数据的校准方法和装置
CN112733879A (zh) 针对不同场景的模型蒸馏方法和装置
CN115809364B (zh) 对象推荐方法和模型训练方法

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2021576943

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2021827170

Country of ref document: EP

Effective date: 20211227

NENP Non-entry into the national phase

Ref country code: DE