WO2022252843A1

WO2022252843A1 - Method and apparatus for training spatio-temporal data processing model, and device and storage medium

Info

Publication number: WO2022252843A1
Application number: PCT/CN2022/086925
Authority: WO
Inventors: 李婷; 包锴楠; 张钧波; 郑宇�
Original assignee: 京东城市(北京)数字科技有限公司
Priority date: 2021-06-02
Filing date: 2022-04-14
Publication date: 2022-12-08
Also published as: CN113723607A

Abstract

Disclosed are a method and apparatus for training a spatio-temporal data processing model, and a device and a storage medium. The method comprises: extracting target spatio-temporal data of any area from a training data set, wherein the training data set comprises spatio-temporal data and attributes of each area; acquiring first reference spatio-temporal data of a first reference area, which has the same attribute as any area, and second reference spatio-temporal data of a second reference area, which has an attribute different from that of any area; inputting the target spatio-temporal data, the first reference spatio-temporal data and the second reference spatio-temporal data into a model to be trained, so as to acquire a first data representation, a second data representation and a third data representation; and if the difference between the first data representation and the second data representation is less than a first threshold value, and the difference between the first data representation and the third data representation is greater than a second threshold value, ending the training of the model.

Description

Spatio-temporal data processing model training method, device, equipment and storage medium

Cross References to Related Applications

This application is based on a Chinese patent application with application number 202110612995.7 and a filing date of June 2, 2021, and claims the priority of this Chinese patent application. The entire content of this Chinese patent application is hereby incorporated by reference into this application.

technical field

The present disclosure relates to the field of computer technology, in particular to the field of artificial intelligence technology such as deep learning and big data, and in particular to a training method, device, equipment and storage medium for a spatio-temporal data processing model.

Background technique

With the development of the mobile Internet, location-based services widely exist in our lives, and a large amount of spatio-temporal data is generated all the time, such as positioning data, mobile phone signaling data, wireless network data, etc. However, accurate and high-quality spatio-temporal data are usually a small amount, and it is difficult and expensive to obtain.

Accurate spatiotemporal data restoration is a very basic function in smart city applications. It can provide support for fine-grained passenger flow inference in scenic spots, offline intelligent marketing, intelligent traffic optimization scheduling, crowd gathering warning, and intelligent location selection. In related technologies, directly using a small amount of high-quality data to train the model may easily lead to over-fitting of the model, which cannot accurately reflect the real data situation. Therefore, how to improve the accuracy and reliability of spatio-temporal data is an urgent problem to be solved.

Contents of the invention

The disclosure provides a training method, device, equipment and storage medium of a spatio-temporal data processing model.

According to a first aspect of the present disclosure, a training method for a spatio-temporal data processing model is provided, including:

Extracting target spatio-temporal data associated with any region from the training data set, wherein the training data set includes spatio-temporal data associated with each region and attribute information of each region;

Obtain from the training data set the first reference spatio-temporal data associated with a first reference area with the same attribute information of any area, and the second reference area associated with a second reference area with different attribute information of any area. Reference spatiotemporal data;

Input the target spatio-temporal data, the first reference spatio-temporal data and the second reference spatio-temporal data into the spatio-temporal data processing model to be trained respectively to obtain the first data representation, the second data representation and the third data representation;

A first degree of difference between the first data representation and the second data representation is less than a first threshold, and a second degree of difference between the first data representation and the third data representation is greater than a second threshold case, end the training process of the spatio-temporal data processing model.

According to a second aspect of the present disclosure, a training device for a spatio-temporal data processing model is provided, including:

An extraction module, configured to extract target spatio-temporal data associated with any region from the training data set, wherein the training data set includes spatio-temporal data associated with each region and attribute information of each region;

The first acquisition module is configured to acquire, from the training data set, first reference space-time data associated with a first reference area that is the same as the attribute information of any area, and a first reference space-time data that is different from the attribute information of any area. Second reference spatio-temporal data associated with the second reference area;

The second acquisition module is configured to respectively input the target spatio-temporal data, the first reference spatio-temporal data and the second reference spatio-temporal data into the spatio-temporal data processing model to be trained, so as to respectively acquire the first data representation, the second reference spatio-temporal data Two data representations and a third data representation;

A judging module, configured to have a first degree of difference between the first data representation and the second data representation less than a first threshold, and a second degree of difference between the first data representation and the third data representation If it is greater than the second threshold, the training process of the spatio-temporal data processing model ends.

According to a third aspect of the present disclosure, an electronic device is provided, including:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor can execute the method described in the above one aspect embodiment.

According to a fourth aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, the computer instructions are used to make the computer execute the method described in the above-mentioned one embodiment.

According to a fifth aspect of the present disclosure, there is provided a computer program product, including a computer program, when the computer program is executed by a processor, the method described in the above one embodiment is implemented.

According to a sixth aspect of the present disclosure, a computer program is provided, the computer program includes computer program code, and when the computer program code is run on a computer, the computer executes the method described in the above one embodiment.

The training method, device, equipment and storage medium of the spatio-temporal data processing model provided by the present disclosure at least have the following

Beneficial effect:

In the embodiment of the present disclosure, the device first extracts the target spatio-temporal data associated with any region from the training data set, wherein the training data set includes the spatio-temporal data associated with each region and the attribute information of each region, and then obtains the target spatio-temporal data associated with any region from the training data set The first reference spatio-temporal data associated with the first reference area with the same attribute information of the area, and the second reference spatio-temporal data associated with the second reference area with different attribute information of any area, and then the target spatio-temporal data, the first reference spatio-temporal data The data and the second reference spatio-temporal data are respectively input into the spatio-temporal data processing model to be trained, so as to obtain the first data representation, the second data representation and the third data representation respectively. Finally, when the first degree of difference between the first data representation and the second data representation is less than the first threshold, and the second degree of difference between the first data representation and the third data representation is greater than the second threshold, the spatio-temporal data processing ends The training process of the model. Therefore, by using comparative self-supervised learning of spatio-temporal data, a spatio-temporal data processing model can be generated, so that spatio-temporal data can be processed to obtain accurate and reliable spatio-temporal data representations, and to improve The accuracy of spatiotemporal prediction provides the conditions.

It should be understood that what is described in this section is not intended to identify key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will be readily understood through the following description.

Description of drawings

The accompanying drawings are used to better understand the present solution, and do not constitute a limitation to the present disclosure. in:

FIG. 1 is a schematic flowchart of a training method for a spatio-temporal data processing model provided according to an embodiment of the present disclosure;

FIG. 2 is a schematic flowchart of a training method for a spatio-temporal data processing model provided according to another embodiment of the present disclosure;

Fig. 3 is a schematic flowchart of a training method for a spatio-temporal data processing model provided according to another embodiment of the present disclosure;

FIG. 4 is a framework diagram of model pre-training of a spatio-temporal data processing model training method provided according to an embodiment of the present disclosure;

FIG. 5 is a framework diagram of population inference for multi-source data fusion provided by an embodiment of the present disclosure;

FIG. 6 is a structural block diagram of a training device for a spatio-temporal data processing model provided by an embodiment of the present disclosure;

Fig. 7 is a structural block diagram of an electronic device provided by an embodiment of the present disclosure.

Detailed ways

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and they should be regarded as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The training method of the spatio-temporal data processing model proposed in this disclosure can be executed by the training device based on the spatio-temporal data processing model provided by this disclosure, and can also be executed by the electronic equipment provided by this disclosure, wherein the electronic equipment can include but not limited to desktop computers, tablet computers and other terminal equipment, which can also be a server. The training device for a spatio-temporal data processing model provided by the present disclosure is used to execute the training method of a spatio-temporal data processing model provided by the present disclosure, which is not a limitation of the present disclosure, hereinafter referred to as for "device".

It should be noted that all user data in this disclosure are obtained in compliance with relevant laws and regulations.

A training method, device, computer equipment, and storage medium for a spatio-temporal data processing model provided by the present disclosure will be described in detail below with reference to the accompanying drawings.

FIG. 1 is a schematic flowchart of a training method for a spatio-temporal data processing model according to an embodiment of the present disclosure.

As shown in Figure 1, the training method of the spatio-temporal data processing model may include the following steps:

Step S101 , extracting target spatio-temporal data associated with any region from the training data set, wherein the training data set includes spatio-temporal data associated with each region and attribute information of each region.

It should be noted that spatio-temporal data has temporal features and spatial features. Temporal features represent the features of spatio-temporal data in the time dimension, and spatial features represent the features of spatio-temporal data in spatial features. In the embodiments of the present disclosure, spatio-temporal data may be real-time positioning data, communication data, real-time population distribution data in an area, etc., which are not limited here.

Wherein, the target spatio-temporal data is the spatio-temporal data associated with any area, where any area may be a scenic area, a school, a shopping mall, etc., which is not limited here.

Wherein, the attribute information of each area may be the location, area, flow of people, population data, traffic conditions, type, etc. of each area, which is not limited here.

Step S102, acquiring first reference space-time data associated with a first reference area with the same attribute information of any area and second reference space-time data associated with a second reference area with different attribute information of any area from the training data set .

It should be noted that, in order to better guarantee the spatio-temporal data processing model, a small amount of data can be used for training to learn the differences between spatio-temporal data corresponding to regions with different attribute information. In some embodiments, the device may obtain from the training dataset the first reference spatio-temporal data associated with the first reference region that has the same attribute information of any region, and acquire the attribute information of any region from the training dataset The second reference spatio-temporal data associated with the second reference area that are different in all of them.

In some embodiments, the first reference spatio-temporal data associated with the first reference region that is the same as at least one attribute information of the attribute information of any region can also be obtained from the training data set, and the attribute information of any region can be obtained from the training data set The second reference spatio-temporal data associated with the second reference area in which each attribute information is different, which is not limited.

In some embodiments, the apparatus may determine the first threshold according to a first matching degree between the attribute information of any region and the attribute information of the first reference region. It can be understood that since any region has the same attribute information as the first reference region, that is, any region and the first reference region should be the same or similar, so the first matching degree between any region and the first reference region is relatively high, The degree of difference is small, so the first threshold is small.

In some embodiments, the apparatus may determine the second threshold according to a second matching degree between the attribute information of any region and the attribute information of the second reference region. It can be understood that, since the attribute information of any area is different from that of the second reference area, that is, the difference between any area and the first reference area should be different or relatively large, so the second matching degree comparison between any area and the second reference area Low, the degree of difference is large, so the second threshold is large.

Step S103, input the target spatio-temporal data, the first reference spatio-temporal data and the second reference spatio-temporal data respectively into the spatio-temporal data processing model to be trained to obtain the first data representation, the second data representation and the third data representation respectively.

It can be understood that the first data representation may represent features of the target spatio-temporal data, which may be vectors, matrices, etc., which is not limited in the present disclosure. Correspondingly, the second data representation may represent the first reference spatio-temporal data, and the third data representation may represent the second reference spatio-temporal data.

Step S104, when the first degree of difference between the first data representation and the second data representation is smaller than the first threshold, and the second difference between the first data representation and the third data representation is greater than the second threshold, end the spatio-temporal The training process of the data processing model.

After acquiring the first data representation, the second data representation and the third data representation, the device may calculate the degree of difference between the first data representation and the second data representation, and between the first data representation and the third data representation. In some embodiments, the apparatus may determine the first degree of difference by comparing the first data representation with the second data representation, and determine the second degree of difference by comparing the first data representation with the third data representation.

Wherein, the first threshold is the threshold of the first degree of difference determined according to the first degree of matching, and the second threshold is the threshold of the second degree of difference determined according to the second degree of matching.

It can be understood that since the first reference area is the same area as the attribute information of any area, the corresponding data representations of the two will be relatively similar. Correspondingly, the second reference area is an area different from the attribute information of any area. As a result, the corresponding data representations of the two will be quite different. And if the first difference degree is less than the first threshold value, it means that the difference between the first reference area and any area is relatively small; if the second difference degree is greater than the second threshold value, it means that the difference between any area and the second reference area The difference is relatively large. Therefore, it can be shown that the data representation obtained by using the current spatio-temporal data processing model is reasonable and reliable, that is, it can meet the needs of the device for processing any spatio-temporal region, so the training process of the spatio-temporal data processing model can be ended.

Fig. 2 is a schematic flowchart of a training method for a spatio-temporal data processing model according to another embodiment of the present disclosure.

As shown in Figure 2, the training method of the spatio-temporal data processing model may include the following steps:

Step S201, extracting target spatio-temporal data associated with any region from the training data set, wherein the training data set includes spatio-temporal data associated with each region and attribute information of each region.

It should be noted that, for the specific implementation process of step S201, reference may be made to the detailed description of any embodiment of the present disclosure, and details are not repeated here.

Step S202, acquiring first reference spatio-temporal data associated with a first reference region that is identical to at least one attribute information of any region's attribute information from the training data set.

In some embodiments, at least one attribute information of the attribute information of the first reference area is the same as that of any area. The first reference spatio-temporal data associated with the area; in some embodiments, the device can also acquire the first reference spatio-temporal data associated with the first reference area that has the same area, traffic status, and population gender distribution in any area, where Not limited.

Step S203, acquiring second reference spatio-temporal data associated with a second reference region whose attribute information of any region is different from each attribute information from the training data set.

It should be noted that, in order to better ensure the spatio-temporal data processing model, a small amount of data can be used for training to learn the differences between the spatio-temporal data corresponding to regions with different attribute information. In this disclosure, the first reference region is attribute information In a case where at least one piece of attribute information is the same as the attribute information of any area, the apparatus may use an area that is different from each attribute information in the attribute information of any area as the second reference area.

In some embodiments, in the present disclosure, the first reference spatio-temporal data associated with the first reference region that has the same attribute information of any region can also be obtained from the training data set; Second reference spatio-temporal data associated with a second reference area with at least one attribute information different among each attribute information of an area.

Step S204, input the target spatio-temporal data, the first reference spatio-temporal data and the second reference spatio-temporal data respectively into the spatio-temporal data processing model to be trained to obtain the first data representation, the second data representation and the third data representation respectively.

Steps It should be noted that, the specific implementation process of step S204 can refer to the above-mentioned embodiment, which will not be repeated here.

Step S205, the first degree of difference between the first data representation and the second data representation is greater than or equal to the first threshold, or the second difference between the first data representation and the third data representation is less than or equal to the second threshold In this case, return to the operation of extracting the target spatio-temporal data from the training data set until the first difference between the first data representation newly generated by the spatio-temporal data processing model to be trained and the newly generated second data representation is less than the first threshold , and when the second degree of difference between the first data representation and the newly generated third data representation is greater than the second threshold, the training process of the spatio-temporal data processing model ends.

It should be noted that if the first degree of difference between the first data representation and the second data representation is greater than or equal to the first threshold, it means that the spatio-temporal data representation of any region obtained after processing by the spatio-temporal data processing model and the first reference region The difference between them is relatively large, and since the attribute information of any region is the same as that of the first reference region, that is, the corresponding spatiotemporal data representations of the two should be the same or similar, which means that the spatiotemporal data processing model at this time is inaccurate and cannot If the condition for ending the training of the neural network training model is met, the device will continue to return to the operation of extracting target spatio-temporal data from the training data set.

In some embodiments, if the second degree of difference between the first data representation and the third data representation is less than or equal to the second threshold, it means that the spatio-temporal data representation of any region obtained after processing by the spatio-temporal data processing model and the second reference region The difference between them is small, and because the attribute information of any region is different from the second reference region, that is, the corresponding spatiotemporal data representations of the two should be different or have a large difference, which can explain that the spatiotemporal data processing model at this time is inaccurate , the condition for ending the training of the neural network training model cannot be met, then the device will continue to return to the operation of extracting the target spatiotemporal data from the training data set.

Further, after returning to the operation of extracting the target spatio-temporal data from the training data set, if the first difference between the newly generated first data representation and the newly generated second data representation of the spatio-temporal data processing model to be trained is less than the first threshold, and the second degree of difference between the first data representation and the newly generated third data representation is greater than the second threshold, then the training process of the spatio-temporal data processing model may end.

In the embodiment of the present disclosure, the device first extracts the target spatio-temporal data associated with any region from the training data set, wherein the training data set includes the spatio-temporal data associated with each region and the attribute information of each region, and then obtains the target spatio-temporal data associated with any region from the training data set The first reference spatio-temporal data associated with at least one of the attribute information of the region is the same as the first reference region, and the first reference region associated with the second reference region that is different in the attribute information of any region is obtained from the training data set. 2. Reference spatio-temporal data, and then input the target spatio-temporal data, the first reference spatio-temporal data and the second reference spatio-temporal data into the spatio-temporal data processing model to be trained to obtain the first data representation, the second data representation and the third data respectively representation, finally the first degree of difference between the first data representation and the second data representation is greater than or equal to the first threshold, or the second difference between the first data representation and the third data representation is less than or equal to the second threshold In this case, return to the operation of extracting the target spatio-temporal data from the training data set until the first difference between the first data representation newly generated by the spatio-temporal data processing model to be trained and the newly generated second data representation is less than the first threshold , and when the second degree of difference between the first data representation and the newly generated third data representation is greater than the second threshold, the training process of the spatio-temporal data processing model ends. Therefore, using comparative self-supervised learning, a large amount of low-quality data can be used to assist in training the spatio-temporal data processing model, so that the time data processing model generated by training can be used to process low-quality spatio-temporal data to obtain more accurate and reliable spatiotemporal data representation.

Fig. 3 is a schematic flowchart of a training method for a spatio-temporal data processing model according to yet another embodiment of the present disclosure.

As shown in Figure 3, the training method of the spatio-temporal data processing model may include the following steps:

Step S301 , extracting target spatio-temporal data associated with any region from the training data set, wherein the training data set includes spatio-temporal data associated with each region and attribute information of each region.

In some embodiments, the spatio-temporal data associated with each area may include a communication data sequence, attribute map, population attribute, and positioning data sequence corresponding to each area, which is not limited.

It should be noted that the attribute map can be constructed by the device based on the known attribute information of the region. For example, the region can be used as a node, the distance of the road network can be used as the side length, and the attribute information of the region such as house price, traffic flow This is not limited as an attribute of each node. The population attribute may be obtained by the device based on the analysis of the acquired positioning data set and communication data, for example, it may be population age distribution, population gender distribution, and the like. This is not limited.

In some embodiments, the training data set may also include labels corresponding to each area, such as the traffic conditions of each area, whether an aggregation event occurs, etc., which is not limited.

Step S302, acquiring first reference spatio-temporal data associated with a first reference area with the same attribute information of any area and second reference spatio-temporal data associated with a second reference area with different attribute information of any area from the training data set .

It should be noted that, the specific implementation process of step S302 may refer to any of the foregoing embodiments, and details are not described here.

Step S303, use the knowledge fusion network to be trained to fuse the communication data sequence, attribute map, population attribute and positioning data sequence corresponding to any region, the first reference region and the second reference region respectively, so as to obtain the fused target Spatio-temporal data, fused first reference spatio-temporal data and second reference spatio-temporal data.

In this disclosure, the knowledge fusion network to be trained can be used to use the knowledge to be trained to use the data such as the communication data sequence, attribute map, population attribute, and positioning data sequence corresponding to any area, the first reference area, and the second reference area. The fusion network performs fusion, so that the fused target spatio-temporal data, the first reference spatio-temporal data, and the second reference spatio-temporal data have spatio-temporal data and attribute information of corresponding regions.

In some embodiments, in this disclosure, the Swapped Contrastive Loss mechanism can be used to predict the spatiotemporal data representation of another region from the spatiotemporal data representation of one region, so that smaller batch data can be used To train the network and reduce the memory requirements during model training.

As shown in FIG. 4 , the frame diagram of model pre-training shown in FIG. 4 can be adopted in the present disclosure, and the model can be trained by using the exchange contrast loss mechanism. As shown in Figure 4, the training framework of the spatio-temporal data processing model can be divided into two types of networks, one is the prototype network (prototype) network, which is a full-connection network (full-connection) network, which mainly undertakes the task of representation calculation , and share network parameters when computing target data and augmented data. The other is the backbone network, which mainly completes the tasks of multi-source data fusion and feature extraction.

It should be noted that the backbone network corresponds to the spatio-temporal knowledge fusion network in Figure 4, and the prototype network corresponds to the encoding network in Figure 4.

In some embodiments, swapped contrastive loss can be used to optimize the parameters of the network, and its expression can be:

L(o _t ,o _s )＝L(o _t ,q _s )+L(q _t ,o _s )

Among them, o _t and o _s can be the output of the backbone network, q _t and q _s can be the output of the prototype network (prototype), and L is the cross-entropy loss function.

It should be noted that L(o _t ,q _s ) can be calculated by the following formula:

Among them, τ is the temperature parameter, and k is the number of protonet networks.

For the convenience of description, for example, take i as the node corresponding to any area, use the side d _{i, j} to represent the distance between the adjacent node j and any area, and use P _a and P _c as the nodes corresponding to each area The two intrinsic features of , among them, each P _a contains the area, traffic conditions and location of the region, and each P _c contains the regional population portrait, such as age and gender, and the reporting point data of dynamic mobile phone positioning

The flow of people in the reaction area is not limited.

Further, to encode P _a and P _c , the specific formula is as follows:

in,

Indicates the input of the i-th node in the network,

and

Represent the input of two embedding (embedding) networks f _a and f _c respectively, and θ ^a and θ ^c represent the parameters of the two embedding networks respectively.

Afterwards, representation learning can be performed using the Message Passing Neural Network (MPNN), where the edge network can be expressed as:

e _ij = f _e (w _ij ; θ ^e )

Among them, w _ij is the input of the edge network, θ ^e is the parameter to be learned, and finally the message passing network can be defined as:

Among them, θ _g is the parameter of the readout (reading) network.

By using the exchange-comparison loss mechanism above, network parameters of the knowledge fusion network to be trained can be optimized, thereby obtaining more accurate target spatio-temporal data, first reference spatio-temporal data, and second reference spatio-temporal data.

It should be noted that the above examples are only for illustrative purposes, and the present disclosure does not limit them.

Step S304, respectively input the fused target spatio-temporal data, fused first reference spatio-temporal data and second reference spatio-temporal data into the spatio-temporal data processing model to be trained, so as to respectively obtain the first data representation, the second data representation Representation and third data representation.

Step S305, when the first degree of difference between the first data representation and the second data representation is smaller than the first threshold, and the second difference between the first data representation and the third data representation is greater than the second threshold, end the spatio-temporal The training process of the data processing model.

It should be noted that, the specific implementation process of S304 and S305 may refer to any of the foregoing embodiments, and details are not described here.

Step S306, input the spatio-temporal data associated with each region into the spatio-temporal data processing model generated by training, so as to obtain the data representation corresponding to each region.

It should be noted that after the training process of the spatio-temporal data processing model is completed, the device can input the spatio-temporal data associated with each region into the spatio-temporal data processing model generated by training, so as to obtain the data representation corresponding to each region. It can be understood that the data representation may represent the spatio-temporal data associated with each region in the form of a static graph structure or in the form of a vector, which is not limited.

As an example, Figure 5 shows a population inference frame diagram of multi-source data fusion. In the above-mentioned embodiment, before obtaining the spatio-temporal data and attribute information of each area, it is necessary to carry out the spatio-temporal characteristics of various types of data in each area Extraction and construction to obtain spatio-temporal data such as communication data sequences, attribute maps, population attributes, and positioning data sequences corresponding to each region. As shown in Figure 5, various types of data can be e-commerce data, travel data, rental data, road network data, and point of interest (POI) data. The extraction and construction of spatio-temporal features to obtain spatio-temporal data such as attribute maps, communication data sequences, and positioning data sequences. Afterwards, by comparing self-supervised model training, using the spatio-temporal deep knowledge fusion network, that is, the knowledge fusion network in the above embodiment, the communication data sequence, attribute map, and population corresponding to any area, the first reference area, and the second reference area are respectively The attributes and positioning data sequences are fused, and then the above-mentioned spatio-temporal data processing model is used to take each fused data as input to obtain each data representation. Finally, the device can input each data representation into the space-time predictor, so as to obtain the predicted flow of people in each area, that is, the inferred value.

Step S307, input the data representation corresponding to each region into the prediction model to be trained, so as to obtain the prediction label corresponding to each region.

It should be noted that the prediction label can be a prediction label for the attribute information of each area, such as the prediction of the flow of people, the prediction of the traffic condition, the prediction of the gender distribution, etc., or it can also be the prediction of the crowd gathering situation, marketing decision, traffic Scheduling decisions and other information are not limited.

It can be understood that for different specific tasks, the corresponding prediction labels obtained by the device through the prediction model may also be different. flow, which is not limited.

In step S308, the prediction model and the spatio-temporal data processing model generated by training are corrected respectively according to the difference between the predicted label and the labeled label.

In some embodiments, the device can compare the predicted label with the labeled label to determine the difference between the predicted label and the labeled label. For example, gradient descent, stochastic gradient descent, etc. can be used to determine the corrected gradient. The prediction model and the spatio-temporal data processing model generated by training are respectively corrected, and the method of determining the correction gradient is not limited in the present disclosure.

In some embodiments, the regression network can also be used to modify the parameters of the spatio-temporal data processing model through the cross loss of the knowledge fusion network.

In some embodiments, the previously trained backbone network and another regression network to be trained can be used to fine-tune and infer specific tasks, and the specific formula is as follows:

Among them, o _t is the output of the backbone network, θ _o is the parameter of the regression network f _o , and σ is the sigmoid activation function. In order to better train the model, a full training strategy can be used to continue training the backbone network and the regression network together under labeled data, which is formally expressed as:

θ _b ＝{θ ^a ,θ ^c ,θ ^e ,θ ^g ,θ ⁿ ,θ ^s }

Among them, L is the loss function, α is the learning rate, and η is the decay coefficient.

In the embodiment of the present disclosure, the device first extracts the target spatio-temporal data associated with any region from the training data set, wherein the training data set includes the spatio-temporal data associated with each region and the attribute information of each region, and then obtains the target spatio-temporal data associated with any region from the training data set The first reference space-time data associated with the first reference area with the same attribute information of the area, and the second reference space-time data associated with the second reference area with different attribute information of any area, and then using the knowledge fusion network to be trained, the The communication data sequence, attribute map, population attribute and positioning data sequence respectively corresponding to any region, the first reference region and the second reference region are fused to obtain the fused target spatio-temporal data, the fused first reference spatio-temporal data and Second reference spatio-temporal data, when the first degree of difference between the first data representation and the second data representation is less than the first threshold, and the second difference between the first data representation and the third data representation is greater than the second threshold , end the training process of the spatiotemporal data processing model, and then input the spatiotemporal data associated with each region into the spatiotemporal data processing model generated by training to obtain the data representation corresponding to each region, and finally input the data representation corresponding to each region to be A predicted model trained to obtain the corresponding predicted label for each region. Therefore, through the fusion of multi-source spatiotemporal data, the accuracy of the prediction model can be improved, and the fine-tuning of the model can be realized while optimizing the model, realizing the purpose of training with a small amount of data and achieving the effect of approximating the effect of full training data.

In order to realize the above-mentioned embodiments, the embodiments of the present disclosure further propose a training device for a spatio-temporal data processing model. FIG. 6 is a structural block diagram of a training device for a spatio-temporal data processing model provided by an embodiment of the present disclosure.

As shown in FIG. 6 , the training device for the spatio-temporal data processing model includes: an extraction module 610 , a first acquisition module 620 , a second acquisition module 630 and a judgment module 640 .

In some embodiments, the attribute information of the region includes the area, population and type of the region, and the first acquisition module is specifically used for:

Obtaining first reference spatio-temporal data associated with a first reference region that is identical to at least one attribute information in the attribute information of any region from the training data set;

The second reference spatio-temporal data associated with the second reference region whose attribute information is different in the attribute information of any region is obtained from the training data set.

Acquiring first reference spatio-temporal data associated with a first reference region having the same attribute information of any region from the training data set;

Second reference spatio-temporal data associated with a second reference region that is different from at least one attribute information of each attribute information of any region is acquired from the training data set.

In some embodiments, the device further includes:

A first determining module, configured to determine the first threshold according to a first degree of matching between the attribute information of any region and the attribute information of the first reference region;

The second determining module is configured to determine the second threshold according to a second degree of matching between the attribute information of any region and the attribute information of the second reference region.

In some embodiments, the judging module is also used for:

The first degree of difference between the first data representation and the second data representation is greater than or equal to a first threshold, or the second degree of difference between the first data representation and the third data representation is less than or If it is equal to the second threshold, return to the operation of extracting the target spatio-temporal data from the training data set until the gap between the first data representation newly generated by the spatio-temporal data processing model to be trained and the newly generated second data representation When the first degree of difference between the first data representation and the newly generated third data representation is greater than the second threshold, the training process of the spatio-temporal data processing model ends.

In some embodiments, the training data set also includes annotated labels corresponding to each of the regions, and the device further includes:

The third acquisition module is used to input the spatio-temporal data associated with each of the regions into the spatio-temporal data processing model generated by training, so as to obtain the data representation corresponding to each of the regions;

The fourth acquisition module is used to input the data representation corresponding to each of the regions into the prediction model to be trained, so as to obtain the prediction labels corresponding to each of the regions;

The correction module is configured to correct the prediction model and the spatio-temporal data processing model generated by the training according to the difference between the prediction label and the label label.

In some embodiments, the spatio-temporal data associated with each of the regions includes communication data sequences, attribute maps, population attributes and positioning data sequences corresponding to each region, and the second acquisition module is also used to:

Using the knowledge fusion network to be trained, fuse the communication data sequence, attribute map, population attribute and positioning data sequence respectively corresponding to the arbitrary region, the first reference region and the second reference region, so as to obtain the fused target space-time data, fused first reference spatio-temporal data and second reference spatio-temporal data;

Input the fused target spatio-temporal data, fused first reference spatio-temporal data and second reference spatio-temporal data into the spatio-temporal data processing model to be trained respectively to obtain the first data representation and the second data representation and the third data representation.

According to the embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, a computer program product, and a computer program.

According to an embodiment of the present disclosure, an electronic device includes:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor can execute the spatiotemporal data processing model described in the above-mentioned embodiments. training method.

According to an embodiment of the present disclosure, the readable storage medium is a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the spatio-temporal data processing model training method described in the above-mentioned embodiments.

According to an embodiment of the present disclosure, the computer program product includes a computer program, and when the computer program is executed by a processor, the method for training the spatio-temporal data processing model described in the above embodiments is implemented.

According to an embodiment of the present disclosure, the computer program includes computer program code, and when the computer program code is run on the computer, the computer executes the method for training the spatio-temporal data processing model described in the above embodiments. FIG. 7 shows a schematic block diagram of an example electronic device 700 that may be used to implement embodiments of the present disclosure. Electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, 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 disclosure described and/or claimed herein.

As shown in FIG. 7, the device 700 includes a computing unit 701 that can execute according to a computer program stored in a read-only memory (ROM) 702 or loaded from a storage unit 708 into a random-access memory (RAM) 703. Various appropriate actions and treatments. In the RAM 703, various programs and data necessary for the operation of the device 700 can also be stored. The computing unit 701, ROM 702, and RAM 703 are connected to each other through a bus 704. An input/output (I/O) interface 705 is also connected to the bus 704 .

Multiple components in the device 700 are connected to the I/O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a magnetic disk, an optical disk, etc. ; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 709 allows the device 700 to exchange information/data with other devices over a computer network such as the Internet and/or various telecommunication networks.

The computing unit 701 may be various general-purpose and/or special-purpose processing components having processing and computing capabilities. Some examples of computing units 701 include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processing processor (DSP), and any suitable processor, controller, microcontroller, etc. The calculation unit 701 executes various methods and processes described above, such as a training method of a spatio-temporal data processing model. For example, in some embodiments, the method for training a spatio-temporal data processing model may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 708 . In some embodiments, part or all of the computer program may be loaded and/or installed on the device 700 via the ROM 702 and/or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the training method of the spatio-temporal data processing model described above can be performed. Alternatively, in other embodiments, the computing unit 701 may be configured in any other appropriate way (for example, by means of firmware) to execute the training method of the spatio-temporal data processing model.

Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips Implemented in a system of systems (SOC), load programmable logic device (CPLD), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpreted on a programmable system including at least one programmable processor, the programmable processor Can be special-purpose or general-purpose programmable processor, can receive data and instruction from storage system, at least one input device, and at least one output device, and transmit data and instruction to this storage system, this at least one input device, and this at least one output device an output device.

Program codes for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special purpose computer, or other programmable data processing devices, so that the program codes, when executed by the processor or controller, make the functions/functions specified in the flow diagrams and/or block diagrams Action is implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer discs, hard drives, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), optical fiber, compact disk read only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the foregoing.

To provide for interaction with the user, the systems and techniques described herein can be implemented on a computer having a display device (e.g., 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 a trackball) through which a user can provide input to the computer. 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 (e.g., visual feedback, auditory feedback, or tactile feedback); and can be in any form (including Acoustic input, speech input or, tactile input) to receive input from the user.

The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., as a a user computer having a graphical user interface or web browser through which a user can interact with embodiments 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 can 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), the Internet, and blockchain networks.

A computer system may include clients and servers. Clients and servers are generally remote from each other and typically 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. The server can be a cloud server, also known as cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the problem of traditional physical host and VPS service ("Virtual Private Server", or "VPS") Among them, there are defects such as difficult management and weak business scalability. The server can also be a server of a distributed system, or a server combined with a blockchain.

It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, each step described in the present disclosure may be executed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution disclosed in the present disclosure can be achieved, no limitation is imposed herein.

The specific implementation manners described above do not limit the protection scope of the present disclosure. It should be apparent to those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made depending on design requirements and other factors. Any modifications, equivalent replacements and improvements made within the principles of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

A training method for a spatio-temporal data processing model, comprising:

Extracting target spatio-temporal data associated with any region from the training data set, wherein the training data set includes spatio-temporal data associated with each region and attribute information of each region;

Obtain from the training data set the first reference spatio-temporal data associated with a first reference area with the same attribute information of any area, and the second reference area associated with a second reference area with different attribute information of any area. Reference spatiotemporal data;

Input the target spatio-temporal data, the first reference spatio-temporal data and the second reference spatio-temporal data into the spatio-temporal data processing model to be trained respectively to obtain the first data representation, the second data representation and the third data representation;

A first degree of difference between the first data representation and the second data representation is less than a first threshold, and a second degree of difference between the first data representation and the third data representation is greater than a second threshold case, end the training process of the spatio-temporal data processing model.
The method according to claim 1, wherein the attribute information of the region includes the area, population and type of the region, and the attribute information of any region obtained from the training data set is The first reference spatio-temporal data associated with the same first reference area, and the second reference spatio-temporal data associated with a second reference area that is different from the attribute information of any area, include:

Obtaining first reference spatio-temporal data associated with a first reference region that is identical to at least one attribute information in the attribute information of any region from the training data set;

The second reference spatio-temporal data associated with the second reference region whose attribute information is different in the attribute information of any region is obtained from the training data set.
The method according to claim 1, wherein the attribute information of the region includes the area, population and type of the region, and the attribute information of any region obtained from the training data set is The first reference spatio-temporal data associated with the same first reference area, and the second reference spatio-temporal data associated with a second reference area that is different from the attribute information of any area, include:

Acquiring first reference spatio-temporal data associated with a first reference region having the same attribute information of any region from the training data set;

Second reference spatio-temporal data associated with a second reference region that is different from at least one attribute information of each attribute information of any region is acquired from the training data set.
The method of claim 3, further comprising:

determining the first threshold according to a first matching degree between the attribute information of any region and the attribute information of the first reference region;

The first threshold is determined according to a second matching degree between the attribute information of any region and the attribute information of the second reference region.
The method according to any one of claims 1 to 4, wherein, after said obtaining the first data representation, the second data representation and the third data representation, further comprising:

The first degree of difference between the first data representation and the second data representation is greater than or equal to a first threshold, or the second degree of difference between the first data representation and the third data representation is less than or If it is equal to the second threshold, return to the operation of extracting the target spatio-temporal data from the training data set until the gap between the first data representation newly generated by the spatio-temporal data processing model to be trained and the newly generated second data representation When the first degree of difference between the first data representation and the newly generated third data representation is greater than the second threshold, the training process of the spatio-temporal data processing model ends.
The method according to any one of claims 1 to 5, wherein the training data set also includes labels corresponding to each of the regions, and the method also includes:

Input the spatio-temporal data associated with each of the regions into the spatio-temporal data processing model generated by training, so as to obtain the data representation corresponding to each of the regions;

Inputting the data representation corresponding to each of the regions into the prediction model to be trained, so as to obtain the prediction labels corresponding to each of the regions;

The prediction model and the spatio-temporal data processing model generated by the training are respectively corrected according to the difference between the prediction label and the annotation label.
The method according to claim 6, wherein the spatio-temporal data associated with each of the regions includes communication data sequences, attribute maps, population attributes, and positioning data sequences corresponding to each region, and the target spatio-temporal The data, the first reference spatiotemporal data and the second reference spatiotemporal data are respectively input into the spatiotemporal data processing model to be trained, including:

Using the knowledge fusion network to be trained, fuse the communication data sequence, attribute map, population attribute and positioning data sequence respectively corresponding to the arbitrary region, the first reference region and the second reference region, so as to obtain the fused target space-time data, fused first reference spatio-temporal data and second reference spatio-temporal data;

Input the fused target spatio-temporal data, fused first reference spatio-temporal data and second reference spatio-temporal data into the spatio-temporal data processing model to be trained respectively to obtain the first data representation and the second data representation and the third data representation.
A training device for a spatio-temporal data processing model, characterized in that it comprises:

An extraction module, configured to extract target spatio-temporal data associated with any region from the training data set, wherein the training data set includes spatio-temporal data associated with each region and attribute information of each region;

The first acquisition module is configured to acquire, from the training data set, first reference space-time data associated with a first reference area that is the same as the attribute information of any area, and a first reference space-time data that is different from the attribute information of any area. Second reference spatio-temporal data associated with the second reference area;

The second acquisition module is configured to respectively input the target spatio-temporal data, the first reference spatio-temporal data and the second reference spatio-temporal data into the spatio-temporal data processing model to be trained, so as to respectively acquire the first data representation, the second reference spatio-temporal data Two data representations and a third data representation;

A judging module, configured to have a first degree of difference between the first data representation and the second data representation less than a first threshold, and a second degree of difference between the first data representation and the third data representation If it is greater than the second threshold, the training process of the spatio-temporal data processing model ends.
The device according to claim 8, wherein the attribute information of the region includes the area, population and type of the region, and the first acquisition module is specifically used for:

Obtaining first reference spatio-temporal data associated with a first reference region that is identical to at least one attribute information in the attribute information of any region from the training data set;

The second reference spatio-temporal data associated with the second reference region whose attribute information is different in the attribute information of any region is obtained from the training data set.
The device according to claim 8, wherein the attribute information of the region includes the area, population and type of the region, and the first acquisition module is specifically used for:

Acquiring first reference spatio-temporal data associated with a first reference region having the same attribute information of any region from the training data set;

Second reference spatio-temporal data associated with a second reference region that is different from at least one attribute information of each attribute information of any region is acquired from the training data set.
The device of claim 10, further comprising:

A first determining module, configured to determine the first threshold according to a first degree of matching between the attribute information of any region and the attribute information of the first reference region;

The second determining module is configured to determine the second threshold according to a second degree of matching between the attribute information of any region and the attribute information of the second reference region.
The device according to any one of claims 8 to 11, wherein the judging module is also used for:

The first degree of difference between the first data representation and the second data representation is greater than or equal to a first threshold, or the second degree of difference between the first data representation and the third data representation is less than or If it is equal to the second threshold, return to the operation of extracting the target spatio-temporal data from the training data set until the gap between the first data representation newly generated by the spatio-temporal data processing model to be trained and the newly generated second data representation When the first degree of difference between the first data representation and the newly generated third data representation is greater than the second threshold, the training process of the spatio-temporal data processing model ends.
The device according to any one of claims 8 to 12, wherein the training data set also includes labels corresponding to each of the regions, and the device also includes:

The third acquisition module is used to input the spatio-temporal data associated with each of the regions into the spatio-temporal data processing model generated by training, so as to obtain the data representation corresponding to each of the regions;

The fourth acquisition module is used to input the data representation corresponding to each of the regions into the prediction model to be trained, so as to obtain the prediction labels corresponding to each of the regions;

The correction module is configured to correct the prediction model and the spatio-temporal data processing model generated by the training according to the difference between the prediction label and the label label.
The device according to claim 13, wherein the spatio-temporal data associated with each of the regions includes communication data sequences, attribute maps, population attributes and positioning data sequences corresponding to each region, and the second acquisition module, Also used for:

Using the knowledge fusion network to be trained, fuse the communication data sequence, attribute map, population attribute and positioning data sequence respectively corresponding to the arbitrary region, the first reference region and the second reference region, so as to obtain the fused target space-time data, fused first reference spatio-temporal data and second reference spatio-temporal data;

Input the fused target spatio-temporal data, fused first reference spatio-temporal data and second reference spatio-temporal data into the spatio-temporal data processing model to be trained respectively to obtain the first data representation and the second data representation and the third data representation.
An electronic device comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor, the instructions are executed by the at least one processor, so that the at least one processor can perform any one of claims 1-7. Methods.
A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.
A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-7.
A computer program, the computer program comprising computer program code, when the computer program code is run on a computer, so that the computer executes the method according to any one of claims 1-7.