CN115759350A - Population mobility prediction method and device for data sparse area - Google Patents

Population mobility prediction method and device for data sparse area Download PDF

Info

Publication number
CN115759350A
CN115759350A CN202211313718.7A CN202211313718A CN115759350A CN 115759350 A CN115759350 A CN 115759350A CN 202211313718 A CN202211313718 A CN 202211313718A CN 115759350 A CN115759350 A CN 115759350A
Authority
CN
China
Prior art keywords
causal
knowledge
data
area
regional
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202211313718.7A
Other languages
Chinese (zh)
Other versions
CN115759350B (en
Inventor
李勇
冯涛
金德鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Original Assignee
Tsinghua University
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 Tsinghua University filed Critical Tsinghua University
Priority to CN202211313718.7A priority Critical patent/CN115759350B/en
Publication of CN115759350A publication Critical patent/CN115759350A/en
Application granted granted Critical
Publication of CN115759350B publication Critical patent/CN115759350B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a population flow prediction method and device for a data sparse area. The method comprises the following steps: acquiring corresponding region causal knowledge from data of a source region by using a causal discovery model based on reinforcement learning; based on the area causal knowledge and an initial variation automatic encoder, obtaining a variation automatic encoder based on causal enhancement, and recovering the missing characteristics of a target area to obtain a potential causal embedded vector; potential causal embedding vectors are characterization vectors corresponding to observed features and missing features; and migrating learning algorithm based on knowledge distillation between the source region and the target region, migrating the knowledge of the prediction model to the target region, and predicting the population flow of the target region based on the knowledge and the embedded vector. The population mobility prediction method for the data sparse area can effectively solve prediction predicament caused by sparse data, and improves population mobility prediction efficiency and accuracy for the data sparse area.

Description

一种数据稀疏区域的人口流动预测方法及装置A population flow prediction method and device in a data-sparse area

技术领域technical field

本发明涉及大数据处理技术领域,具体涉及一种数据稀疏区域的人口流动预测方法及装置。另外,还涉及一种电子设备及处理器可读存储介质。The invention relates to the technical field of big data processing, in particular to a population flow prediction method and device in a data-sparse area. In addition, it also relates to an electronic device and a processor-readable storage medium.

背景技术Background technique

人口流动(Population Mobility)反映了城市的城市结构和人的设施需求分布。准确的人口流动预测可以帮助人们更好地提前了解和规划城市结构和设施需求分布,从而降低人们的出行成本,提高城市的效率。对于发展中城市(即数据稀疏区域),人口流动性预测起着至关重要的作用,因为良好的城市结构和设施需求分布对于发展中城市的结构布局和未来发展具有重要意义。Population Mobility reflects the city's urban structure and the distribution of people's facility needs. Accurate population flow forecasting can help people better understand and plan the urban structure and facility demand distribution in advance, thereby reducing people's travel costs and improving the efficiency of the city. For developing cities (i.e., data-sparse regions), population mobility prediction plays a crucial role, because a good urban structure and facility demand distribution are of great significance for the structural layout and future development of developing cities.

目前,现有技术通常基于简单的物理规律对人口流动进行建模和预测,建模能力有限,无法表达复杂的流动模式。随着机器学习和深度学习的快速发展,基于决策树的模型和图神经网络等复杂模型凸显了其对人口流动性预测的强大能力。尽管如此,这些方法需要大量数据来拟合复杂的模型,因此在发展中城市的应用受到了限制。该方法中的每一种都需要为某个城市建立模型,并使用大量数据拟合模型的参数。所以无法帮助预测发展中城市的人口流动,因为数据收集不足使得一些关键特征无法观察到,导致针对数据稀疏区域的人口流动预测效率和精确度较差。因此,如何设计一种数据稀疏区域的人口流动预测方案来提升人口流动预测效率和精确度成为亟待解决的难题。At present, existing technologies usually model and predict population flow based on simple physical laws, with limited modeling capabilities and unable to express complex flow patterns. With the rapid development of machine learning and deep learning, complex models such as decision tree-based models and graph neural networks have highlighted their powerful ability to predict population mobility. Nonetheless, these methods require large amounts of data to fit complex models, thus limiting their application to developing cities. Each of these methods requires building a model for a city and fitting the parameters of the model with a large amount of data. Therefore, it cannot help predict the population flow of developing cities, because insufficient data collection makes some key features unobservable, resulting in poor efficiency and accuracy of population flow prediction for data-sparse areas. Therefore, how to design a population flow prediction scheme in data-sparse areas to improve the efficiency and accuracy of population flow prediction has become an urgent problem to be solved.

发明内容Contents of the invention

为此,本发明提供一种数据稀疏区域的人口流动预测方法及装置,以解决现有技术中存在的数据稀疏区域的人口流动预测方案局限性较高,从而导致人口流动预测效率和精确度较差的缺陷。For this reason, the present invention provides a method and device for population flow prediction in data-sparse areas to solve the limitations of the population flow prediction scheme in data-sparse areas in the prior art, resulting in relatively low efficiency and accuracy of population flow prediction. poor defect.

第一方面,本发明提供一种数据稀疏区域的人口流动预测方法,包括:In a first aspect, the present invention provides a population flow prediction method in a data-sparse area, including:

利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识;Use a causal discovery model based on reinforcement learning to obtain the corresponding regional causal knowledge from the data of the source region;

基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;其中,所述潜在的因果嵌入向量是观测特征以及所述缺失特征对应的表征向量;Based on the causal knowledge of the region and the initial variational autoencoder, a variational autoencoder based on causal enhancement is obtained; the missing features of the target region are recovered by using the variational autoencoder based on causal enhancement to obtain potential A causal embedding vector; wherein, the potential causal embedding vector is an observation feature and a representation vector corresponding to the missing feature;

基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;其中,所述预测模型是预先基于所述源区域的数据构建的人口流动预测模型。Based on the transfer learning algorithm of knowledge distillation between the source area and the target area, the knowledge of the prediction model is transferred to the target area, and the knowledge of the prediction model and the potential causal embedding vector are used to perform the Population flow prediction in the target area; wherein, the prediction model is a population flow prediction model constructed in advance based on the data of the source area.

进一步的,所述利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识,具体包括:Further, the use of the causal discovery model based on reinforcement learning to obtain the corresponding regional causal knowledge from the data of the source region specifically includes:

获取所述源区域的数据;基于强化学习的区域因果知识构建策略确定基于区域属性特征排序的因果发现模型,利用所述因果发现模型对所述源区域的数据进行分析,以获得满足预设条件的区域属性特征顺序,并通过贝叶斯检验修剪得到包含区域属性特征之间关系的特征因果图;其中,所述特征因果图用于表示所述区域因果知识。Acquiring the data of the source area; determining the causal discovery model based on the ranking of regional attribute characteristics based on the regional causal knowledge construction strategy of reinforcement learning, and using the causal discovery model to analyze the data of the source area to obtain the The sequence of regional attribute features is obtained by Bayesian test pruning to obtain a feature causal graph containing the relationship between regional attribute features; wherein, the feature causal graph is used to represent the regional causal knowledge.

进一步的,利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量,具体包括:Further, the causal enhancement-based variational autoencoder is used to restore the missing features of the target region to obtain a potential causal embedding vector, which specifically includes:

将所述特征因果图作为特征恢复路径,基于所述初始的变分自动编码器和所述特征恢复路径来学习包含区域属性特征的缺失信息,得到所述潜在的因果嵌入向量。The feature causal graph is used as a feature recovery path, and the missing information including region attribute features is learned based on the initial variational autoencoder and the feature recovery path to obtain the potential causal embedding vector.

进一步的,基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器,具体包括:Further, based on the regional causal knowledge and the initial variational autoencoder, a variational autoencoder based on causal enhancement is obtained, specifically including:

将所述目标区域的未观测到的缺失特征显式建模为辅助的潜在变量,并使用所述区域因果知识对应的因果路径将特征之间的关联关系添加到所述初始的变分自动编码器,以构建得到初始的基于因果增强的变分自动编码器;通过反向传播训练因果增强所述初始的基于因果增强的变分自动编码器,得到所述基于因果增强的变分自动编码器。Explicitly model the unobserved missing features of the target region as auxiliary latent variables, and use the causal path corresponding to the causal knowledge of the region to add the association relationship between features to the initial variational autoencoding To construct the initial variational autoencoder based on causal enhancement; through backpropagation training, causally enhance the initial variational autoencoder based on causal enhancement, and obtain the variational autoencoder based on causal enhancement .

进一步的,基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测,具体包括:Further, the population flow prediction of the target area is performed based on the knowledge of the prediction model and the potential causal embedding vector, specifically including:

基于所述预测模型的知识和所述潜在的因果嵌入向量获取所述目标区域中起点区域的特征和终点区域的特征,并添加起点和终点之间的距离信息,来预测所述目标区域中起点和终点之间的人流量。Based on the knowledge of the predictive model and the potential causal embedding vector, obtain the features of the start area and the end area in the target area, and add the distance information between the start point and the end point to predict the start point in the target area traffic between the destination and the destination.

进一步的,所述区域因果知识为源区域中人员特征和区域属性特征之间的因果映射关系信息。Further, the regional causal knowledge is causal mapping relationship information between personnel characteristics and regional attribute characteristics in the source region.

进一步的,所述目标区域为数据稀疏区域。Further, the target area is a data-sparse area.

第二方面,本发明还提供一种数据稀疏区域的人口流动预测装置,包括:In the second aspect, the present invention also provides a device for predicting population flow in a data-sparse area, including:

区域因果知识获取单元,用于利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识;The regional causal knowledge acquisition unit is used to obtain the corresponding regional causal knowledge from the data of the source region by using the causal discovery model based on reinforcement learning;

因果嵌入向量获取单元,用于基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;其中,所述潜在的因果嵌入向量是观测特征以及所述缺失特征对应的表征向量;The causal embedding vector acquisition unit is used to obtain a causal-enhanced variational autoencoder based on the regional causal knowledge and an initial variational autoencoder; using the causal-enhanced variational autoencoder for the target region The missing feature is restored to obtain a potential causal embedding vector; wherein the potential causal embedding vector is an observation feature and a representation vector corresponding to the missing feature;

人口流动预测单元,用于基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;其中,所述预测模型是预先基于所述源区域的数据构建的人口流动预测模型。The population flow prediction unit is used to transfer the knowledge of the prediction model to the target area based on the migration learning algorithm of knowledge distillation between the source area and the target area, and based on the knowledge of the prediction model and the potential The causal embedding vector is used to predict the population flow of the target area; wherein, the prediction model is a population flow prediction model constructed in advance based on the data of the source area.

进一步的,所述区域因果知识获取单元,具体用于:Further, the regional causal knowledge acquisition unit is specifically used for:

获取所述源区域的数据;基于强化学习的区域因果知识构建策略确定基于区域属性特征排序的因果发现模型,利用所述因果发现模型对所述源区域的数据进行分析,以获得满足预设条件的区域属性特征顺序,并通过贝叶斯检验修剪得到包含区域属性特征之间关系的特征因果图;其中,所述特征因果图用于表示所述区域因果知识。Acquiring the data of the source area; determining the causal discovery model based on the ranking of regional attribute characteristics based on the regional causal knowledge construction strategy of reinforcement learning, and using the causal discovery model to analyze the data of the source area to obtain the The sequence of regional attribute features is obtained by Bayesian test pruning to obtain a feature causal graph containing the relationship between regional attribute features; wherein, the feature causal graph is used to represent the regional causal knowledge.

进一步的,所述因果嵌入向量获取单元,具体用于:Further, the causal embedding vector acquisition unit is specifically used for:

将所述特征因果图作为特征恢复路径,基于所述初始的变分自动编码器和所述特征恢复路径来学习包含区域属性特征的缺失信息,得到所述潜在的因果嵌入向量。The feature causal graph is used as a feature recovery path, and the missing information including region attribute features is learned based on the initial variational autoencoder and the feature recovery path to obtain the potential causal embedding vector.

进一步的,所述因果嵌入向量获取单元,具体还用于:Further, the causal embedding vector acquisition unit is also specifically used for:

将所述目标区域的未观测到的缺失特征显式建模为辅助的潜在变量,并使用所述区域因果知识对应的因果路径将特征之间的关联关系添加到所述初始的变分自动编码器,以构建得到初始的基于因果增强的变分自动编码器;通过反向传播训练因果增强所述初始的基于因果增强的变分自动编码器,得到所述基于因果增强的变分自动编码器。Explicitly model the unobserved missing features of the target region as auxiliary latent variables, and use the causal path corresponding to the causal knowledge of the region to add the association relationship between features to the initial variational autoencoding To construct the initial variational autoencoder based on causal enhancement; through backpropagation training, causally enhance the initial variational autoencoder based on causal enhancement, and obtain the variational autoencoder based on causal enhancement .

进一步的,所述人口流动预测单元,具体用于:Further, the population flow prediction unit is specifically used for:

基于所述预测模型的知识和所述潜在的因果嵌入向量获取所述目标区域中起点区域的特征和终点区域的特征,并添加起点和终点之间的距离信息,来预测所述目标区域中起点和终点之间的人流量。Based on the knowledge of the predictive model and the potential causal embedding vector, obtain the features of the start area and the end area in the target area, and add the distance information between the start point and the end point to predict the start point in the target area traffic between the destination and the destination.

进一步的,所述区域因果知识为源区域中人员特征和区域属性特征之间的因果映射关系信息。Further, the regional causal knowledge is causal mapping relationship information between personnel characteristics and regional attribute characteristics in the source region.

进一步的,所述目标区域为数据稀疏区域。Further, the target area is a data-sparse area.

第三方面,本发明还提供一种电子设备,包括:存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序,处理器执行所述计算机程序时实现如上述任意一项所述的数据稀疏区域的人口流动预测方法的步骤。In a third aspect, the present invention also provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and operable on the processor. The steps of the method for predicting population flow in data-sparse regions are described.

第四方面,本发明还提供一种处理器可读存储介质,所述处理器可读存储介质上存储有计算机程序,该计算机程序被处理器执行时实现如上述任意一项所述的数据稀疏区域的人口流动预测方法的步骤。In a fourth aspect, the present invention also provides a processor-readable storage medium, where a computer program is stored on the processor-readable storage medium, and when the computer program is executed by a processor, the data sparseness as described in any one of the above is realized. The steps of the regional population mobility prediction method.

本发明提供的数据稀疏区域的人口流动预测方法,通过基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识,并基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;然后,利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;进而基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;有效解决稀疏数据导致的预测困境,提高了针对数据稀疏区域的人口流动预测效率和精确度。The method for predicting population flow in a data-sparse region provided by the present invention obtains the corresponding regional causal knowledge from the data of the source region through a causal discovery model based on reinforcement learning, and based on the regional causal knowledge and the initial variational autoencoder, Obtain a variational autoencoder based on causal enhancement; then, use the variational autoencoder based on causal enhancement to restore the missing features of the target region to obtain a potential causal embedding vector; then based on the source region and the A migration learning algorithm of knowledge distillation between target areas, which migrates the knowledge of the prediction model to the target area, and performs population flow prediction in the target area based on the knowledge of the prediction model and the potential causal embedding vector; effectively Solve the prediction dilemma caused by sparse data, and improve the efficiency and accuracy of population flow prediction for data-sparse areas.

附图说明Description of drawings

为了更清楚地说明本发明实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作一简单地介绍,显而易见地,下面描述中的附图是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获取其他的附图。In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description These are some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained based on these drawings without creative effort.

图1是本发明实施例提供的数据稀疏区域的人口流动预测方法的流程示意图;FIG. 1 is a schematic flowchart of a population flow prediction method in a data-sparse region provided by an embodiment of the present invention;

图2是本发明实施例提供的基于因果增强的数据稀疏区域的人口流动预测原理示意图;FIG. 2 is a schematic diagram of the principle of population flow prediction based on causally enhanced data-sparse regions provided by an embodiment of the present invention;

图3是本发明实施例提供的特征恢复模型的结构示意图;Fig. 3 is a schematic structural diagram of a feature restoration model provided by an embodiment of the present invention;

图4是本发明实施例提供的基于知识蒸馏的目标区域的预测过程示意图;Fig. 4 is a schematic diagram of a prediction process of a target area based on knowledge distillation provided by an embodiment of the present invention;

图5是本发明实施例提供的数据稀疏区域的人口流动预测方法的具体示意图;FIG. 5 is a specific schematic diagram of a population flow prediction method in a data-sparse area provided by an embodiment of the present invention;

图6是本发明实施例提供的数据稀疏区域的人口流动预测装置的结构示意图;6 is a schematic structural diagram of a population flow prediction device in a data-sparse area provided by an embodiment of the present invention;

图7是本发明实施例提供的电子设备的实体结构示意图。FIG. 7 is a schematic diagram of a physical structure of an electronic device provided by an embodiment of the present invention.

具体实施方式Detailed ways

为使本发明实施例的目的、技术方案和优点更加清楚,下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获取的所有其他实施例,都属于本发明保护的范围。In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

本发明提出了一种数据稀疏区域的人口流动预测方法,基于因果增强变分自动编码器的新型人口流动预测模型,将三种类型的城市知识从源区域(即数据丰富的城市)转移到目标区域(即发展中城市)。具体的,首先利用基于强化学习的因果发现模型来搜索发现数据丰富的城市中的区域因果知识(即城市因果知识),并结合基于VAE的恢复模型得到特征恢复模型(Causal Enhanced Variational Auto-encoder,CEVAE),用于获取相应的缺失特征;然后将获得的特征恢复模型转移到发展中城市以获得缺失的潜在表示特征;最后,通过数据丰富的城市与发展中城市之间的知识蒸馏,利用基于知识蒸馏的迁移学习方将预测模型的知识转移到发展中城市,进行人口流动预测。The present invention proposes a method for population flow prediction in data-sparse regions, based on a novel population flow prediction model based on causally augmented variational autoencoders, to transfer three types of urban knowledge from source regions (i.e., data-rich cities) to target regions (i.e. developing cities). Specifically, firstly, the causal discovery model based on reinforcement learning is used to search and discover the regional causal knowledge in cities with rich data (that is, urban causal knowledge), and combined with the recovery model based on VAE to obtain a feature recovery model (Causal Enhanced Variational Auto-encoder, CEVAE) to obtain the corresponding missing features; then the obtained feature restoration model is transferred to developing cities to obtain missing latent representation features; finally, through knowledge distillation between data-rich cities and developing cities, using the The transfer learning method of knowledge distillation transfers the knowledge of the prediction model to developing cities for population flow prediction.

下面基于本发明所述的数据稀疏区域的人口流动预测方法,对其实施例进行详细描述。如图1所示,其为本发明实施例提供的数据稀疏区域的人口流动预测方法的流程示意图,具体过程包括以下步骤:Embodiments of the method for predicting population flow in data-sparse regions of the present invention will be described in detail below. As shown in Figure 1, it is a schematic flow chart of a method for predicting population flow in a data-sparse area provided by an embodiment of the present invention, and the specific process includes the following steps:

步骤101:利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识。所述区域因果知识为源区域中人员特征和区域属性特征之间的因果映射关系信息,比如若人员特征为公司白领或者商务人士等身份属性,则该人员特征与咖啡厅、快餐店等区域属性特征之间存在因果映射关系。所述因果映射关系信息即为从源区域的数据中提取的人员特征和区域属性特征之间存在的因果映射关系。所述源区域可以是指数据丰富的发达城市。Step 101: Using a causal discovery model based on reinforcement learning to obtain the corresponding regional causal knowledge from the data of the source region. The regional causal knowledge is the causal mapping relationship information between the personnel characteristics and the regional attribute characteristics in the source region. There is a causal mapping between features. The causal mapping relationship information is the causal mapping relationship between the personnel features extracted from the source area data and the area attribute features. The source area may refer to a data-rich developed city.

在本发明实施例中,首先获取所述源区域的数据,然后基于强化学习的区域因果知识构建策略确定基于区域属性特征排序的因果发现模型,并利用所述因果发现模型对所述源区域的数据进行分析,以获得满足预设条件的区域属性特征顺序,并通过贝叶斯检验修剪得到包含区域属性特征之间关系的特征因果图。其中,所述特征因果图用于表示所述区域因果知识(即城市因果知识)。所述因果发现模型是基于观测的特征数据,发现特征存在的因果关系的深度学习模型。In the embodiment of the present invention, first obtain the data of the source region, and then determine the causal discovery model based on the ranking of regional attribute characteristics based on the regional causal knowledge construction strategy of reinforcement learning, and use the causal discovery model to analyze the source regions. The data is analyzed to obtain the sequence of regional attribute features that meet the preset conditions, and the feature causal graph containing the relationship between regional attribute features is obtained through Bayesian test pruning. Wherein, the characteristic causal graph is used to represent the regional causal knowledge (ie city causal knowledge). The causal discovery model is a deep learning model that discovers the causal relationship of features based on observed feature data.

在具体实施过程中,可将城市因果知识构建建模为基于区域属性排序的因果发现问题,得到相应的因果发现模型,并利用强化学习(Reinforcement Learning,RL)来解决该问题。为了将RL纳入基于区域特征(即区域属性)排序的因果发现问题,将区域特征排序搜索问题制定为多步马尔可夫决策过程(MarkovDecisionProcess,MDP)。例如,在形式上,每个MDP模型可以描述为一个4元组(S,A,P,R)。特别地,S和A分别代表状态空间和动作空间。P:S×A→R表示状态转换的概率。即P(st+1|st,at)是下一个状态st+1以当前状态st和动作at为条件的概率分布;最后,R:S×A→R是奖励函数;R(s,a)代表收到的奖励通过在状态,s∈S下执行动作a∈A。In the specific implementation process, the construction of urban causal knowledge can be modeled as a causal discovery problem based on regional attribute ranking, and the corresponding causal discovery model can be obtained, and reinforcement learning (RL) can be used to solve this problem. In order to incorporate RL into the causal discovery problem based on the ranking of region features (i.e., region attributes), the region feature ranking search problem is formulated as a multi-step Markov decision process (MarkovDecisionProcess, MDP). For example, formally, each MDP model can be described as a 4-tuple (S,A,P,R). In particular, S and A denote the state space and action space, respectively. P:S×A→R represents the probability of state transition. That is, P(s t+1 |s t , a t ) is the probability distribution of the next state s t+1 conditional on the current state s t and action a t ; finally, R:S×A→R is the reward function; R(s,a) represents the reward received by performing an action a ∈ A in a state, s ∈ S.

下面详细说明如何对MDP模型的相应组件进行建模:状态(State):现有技术中通过直接使用观察到的区域属性F作为状态来捕捉潜在的因果关系;本发明使用编码器将每个属性数据Fi嵌入到状态si中,这有利于排序搜索过程,因此,可以得到状态空间为

Figure BDA0003908119780000081
并利用Transformer结构中基于自我注意的编码器作为的编码器。动作(Action):动作空间A={Fi|i=1,2,…,|F|}由所有区域属性组成;在每个决策步骤中,选择一个区域属性作为一个动作,属性选择的顺序就是搜索顺序,动作空间大小|F|等于所有区域属性的个数,比其他因果发现方法的2|F|*|F|搜索空间小很多,大大提高了顺序搜索的效率。状态转换(State transition):在的问题中,状态转换是确定性的,并且与当前决策步骤中选择的动作有关;也就是说,如果在当前决策步骤t选择at=Fj,那么下一个状态是st+1=sj=encoder(Fj)。奖励(Reward):在问题中设置了一个情节奖励;情节奖励
Figure BDA0003908119780000082
只有在得到区域属性顺序Π时才能获得,其描述了两者之间的匹配程度因果结构和观察到的区域属性。The following details how to model the corresponding components of the MDP model: State (State): in the prior art, the potential causal relationship is captured by directly using the observed region attribute F as the state; the present invention uses an encoder to convert each attribute The data F i is embedded into the state s i , which facilitates the sorting search process, therefore, the state space can be obtained as
Figure BDA0003908119780000081
And use the self-attention-based encoder in the Transformer structure as the encoder. Action: The action space A={F i |i=1,2,…,|F|} consists of all region attributes; in each decision step, a region attribute is selected as an action, and the order of attribute selection It is the search order. The size of the action space |F| is equal to the number of all region attributes, which is much smaller than the 2 |F|*|F| search space of other causal discovery methods, which greatly improves the efficiency of sequential search. State transition: In the problem of , the state transition is deterministic and related to the action chosen in the current decision step; that is, if a t = F j is chosen at the current decision step t, then the next The state is s t+1 =s j =encoder(F j ). Reward: A plot reward is set in the question; plot reward
Figure BDA0003908119780000082
This is only obtained when the region attribute order Π is obtained, which describes the degree of match between the causal structure and the observed region attributes.

基于上述MDP模型,基于区域属性排序的因果发现由策略函数π:Ω→A描述。特别地,π(a|s)表示在当前状态s下选择动作a的概率。采用基于长短期记忆网络(LSTM,LongShort-Term Memory)的解码器,将状态映射到动作,即at=decoder(st)。基于RL框架,引入了相应的策略梯度来训练基于排序的因果发现策略模型,以获得最佳的区域属性顺序。最佳区域属性顺序对应于完全连接的DAG(有向无环图),并通过贝叶斯检验修剪得到最终的特征因果图

Figure BDA0003908119780000083
Based on the above MDP model, causal discovery based on region attribute ranking is described by a policy function π:Ω→A. In particular, π(a|s) denotes the probability of choosing action a in the current state s. A decoder based on a long-short-term memory network (LSTM, LongShort-Term Memory) is used to map the state to an action, that is, a t =decoder(s t ). Based on the RL framework, a corresponding policy gradient is introduced to train a ranking-based causal discovery policy model for optimal region attribute ordering. The optimal region attribute order corresponds to a fully connected DAG (directed acyclic graph), and is pruned by Bayesian tests to obtain the final feature causal graph
Figure BDA0003908119780000083

基于上述实现过程,最终能够得到包含区域特征之间关系的特征因果图所表示的城市因果知识,以便于后续应用。Based on the above implementation process, the urban causal knowledge represented by the characteristic causal graph including the relationship between regional features can be finally obtained for subsequent application.

步骤102:基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;其中,所述潜在的因果嵌入向量是观测特征以及所述缺失特征对应的表征向量。所述目标区域为数据稀疏区域。所述初始的变分自动编码器是指基于VAE(Variational Auto-encoder)的恢复模型,所述基于因果增强的变分自动编码器是指特征恢复模型,即如图2中的CEVAE,其包括编码器和解码器。Step 102: Based on the causal knowledge of the region and the initial variational autoencoder, obtain a variational autoencoder based on causal enhancement; use the variational autoencoder based on causal enhancement to restore the missing features of the target region, A potential causal embedding vector is obtained; wherein, the potential causal embedding vector is an observation feature and a representation vector corresponding to the missing feature. The target area is a data sparse area. The initial variational autoencoder refers to a recovery model based on VAE (Variational Auto-encoder), and the variational autoencoder based on causal enhancement refers to a feature recovery model, such as CEVAE in Figure 2, which includes encoder and decoder.

在本发明实施例中,可将所述目标区域的未观测到的缺失特征显式建模为辅助的潜在变量,并使用所述区域因果知识对应的因果路径将特征之间的关联关系添加到所述初始的变分自动编码器,以构建得到初始的基于因果增强的变分自动编码器;通过反向传播训练因果增强所述初始的基于因果增强的变分自动编码器,得到所述基于因果增强的变分自动编码器。进一步的,将所述特征因果图作为特征恢复路径,基于所述初始的变分自动编码器和所述特征恢复路径来学习包含区域属性特征的缺失信息,得到所述潜在的因果嵌入向量;所述潜在的因果嵌入向量为基于观测特征恢复的缺失特征的嵌入向量。In the embodiment of the present invention, the unobserved missing features of the target region can be explicitly modeled as auxiliary latent variables, and the causal path corresponding to the causal knowledge of the region is used to add the correlation between features to The initial variational autoencoder is constructed to obtain an initial variational autoencoder based on causal enhancement; the initial variational autoencoder based on causal enhancement is trained through backpropagation to obtain the initial variational autoencoder based on causal enhancement Variational Autoencoders with Causal Augmentation. Further, the feature causal graph is used as a feature recovery path, based on the initial variational autoencoder and the feature recovery path to learn missing information including regional attribute features, and obtain the potential causal embedding vector; The latent causal embedding vector is the embedding vector of missing features based on observed feature recovery.

具体的,基于CEVAE的特征恢复过程包括:将在上述获得的特征因果图

Figure BDA0003908119780000091
作为特征恢复路径来学习包含区域特征的缺失信息,以解决发展中城市的数据稀疏问题。如图3所示,CEVAE的结构基于VAE实现,缺失的特征在编码器和解码器中都建模为潜在变量,以支持潜在的因果嵌入向量的学习Z。图3中得到的潜在的因果嵌入向量Z(或μzz)传递给后续的预测过程。其中,所述潜在的因果嵌入向量对应缺失特征的信息。所述基于因果增强的变分自动编码器包括因果估计增强型编码器。该因果估计增强型编码器与传统的VAE相同,所述CEVAE的编码器学习了一个条件分布q(Z|X),该条件分布通过隐变量Z(即潜在的因果嵌入向量)对部分观察特征X(即观测特征)的依赖性进行建模。为了学习代表性的嵌入向量并解决发展中城市数据稀疏的障碍,本发明还应考虑缺失特征的信息。如图3所示,对缺失特征的潜在变量Y进行建模,这也是Z的因果依赖关系。通过对缺失特征的显式建模,潜在的因果嵌入向量Z的分布取决于观察到的特征X(即观测特征)和缺失特征Y。具体来说,编码器由NY变量估计器组成,用于估计每个缺失特征的分布,每个估计器对估计缺失特征的分布进行建模,其因果变量是根据特征因果图
Figure BDA0003908119780000101
的父节点。每个估计器都是一个参数独立的多层感知器(Multilayer Perceptron,MLP),以相应的因果变量作为输入,估计分布作为输出。将所有变量的分布视为正态分布,例如
Figure BDA0003908119780000102
所以估计器可以描述如下。Specifically, the CEVAE-based feature recovery process includes: the feature causal graph obtained above
Figure BDA0003908119780000091
As a feature recovery path to learn missing information containing regional features to address data sparsity in developing cities. As shown in Fig. 3, the structure of CEVAE is implemented based on VAE, and missing features are modeled as latent variables in both encoder and decoder to support the learning of latent causal embedding vectors Z. The potential causal embedding vector Z (or μ z , σ z ) obtained in Fig. 3 is passed to the subsequent prediction process. Wherein, the latent causal embedding vector corresponds to information of missing features. The causal augmentation based variational autoencoder includes a causal estimation augmented encoder. The causal estimation enhanced encoder is the same as the traditional VAE. The encoder of the CEVAE learns a conditional distribution q(Z|X) that uses the latent variable Z (i.e., the latent causal embedding vector) to partially observe the features X (that is, the observed features) to model the dependence. In order to learn representative embedding vectors and address the obstacle of data sparsity in developing cities, the present invention should also consider the information of missing features. As shown in Figure 3, a latent variable Y with missing features is modeled, which is also a causal dependency of Z. By explicitly modeling missing features, the distribution of latent causal embedding vectors Z depends on observed features X (i.e., observed features) and missing features Y. Specifically, the encoder consists of N Y variable estimators for estimating the distribution of each missing feature, each estimator models the distribution of the estimated missing feature, and its causal variables are based on the feature causal graph
Figure BDA0003908119780000101
the parent node of . Each estimator is a parameter-independent Multilayer Perceptron (MLP) that takes the corresponding causal variable as input and estimates the distribution as output. Treat the distribution of all variables as normal, e.g.
Figure BDA0003908119780000102
So the estimator can be described as follows.

Figure BDA0003908119780000103
Figure BDA0003908119780000103

其中,||表示变量的串联,

Figure BDA0003908119780000104
表示
Figure BDA0003908119780000109
中节点i的父节点,j表示i的父节点;
Figure BDA0003908119780000105
表示高斯分布,μi表示高斯分布的均值以及σi表示高斯分布的方差。很明显,可以观察到某个特定特征的因果变量或缺失特征。因此,缺失的因果变量将由其相应的估计器来估计。假设基础特征,例如人口数量,将很容易获得,因此特征因果图
Figure BDA0003908119780000106
的根节点不会丢失,并且临时路径的起始节点将使后续节点能够被推断。Among them, || represents the concatenation of variables,
Figure BDA0003908119780000104
express
Figure BDA0003908119780000109
The parent node of the middle node i, j represents the parent node of i;
Figure BDA0003908119780000105
denotes a Gaussian distribution, μ i denotes the mean of the Gaussian distribution and σ i denotes the variance of the Gaussian distribution. Clearly, a causal variable or missing feature for a particular feature can be observed. Therefore, missing causal variables will be estimated by their corresponding estimators. Assuming that underlying features, such as population size, will be readily available, the feature causal graph
Figure BDA0003908119780000106
The root node of is not lost, and the starting node of the temporary path will enable subsequent nodes to be inferred.

给定观察到的特征(即观测特征)和估计量,可以通过从估计分布中采样来估计缺失的特征。可以通过结合观察到的特征和估计量来收集完整的信息。因此,潜在变量Z(即潜在的因果嵌入向量)的分布将根据完整信息进行估计,例如P(Z|Y,X)。估计量也是MLP。该过程可以表示如下。Given observed features (i.e. observed features) and estimators, missing features can be estimated by sampling from the estimated distribution. Complete information can be gathered by combining observed features and estimators. Therefore, the distribution of the latent variable Z (i.e., the underlying causal embedding vector) will be estimated from complete information, such as P(Z|Y,X). The estimator is also the MLP. The process can be expressed as follows.

Figure BDA0003908119780000107
Figure BDA0003908119780000107

其中,||表示变量的串联,X表示观察到的特征,

Figure BDA0003908119780000108
表示估计的缺失特征;μ_{Z},σ_{Z}分别表示隐变量z的均值与方差。Among them, || represents the concatenation of variables, X represents the observed features,
Figure BDA0003908119780000108
Represents the estimated missing features; μ_{Z}, σ_{Z} represent the mean and variance of the latent variable z, respectively.

在因果估计增强型解码器中,解码器是基于隐变量Z(即潜在变量Z)恢复或重建包括观察到的特征和未观察到的特征的完整信息,例如p(X|Z)和p(Y|X,Z)。利用特征因果图

Figure BDA0003908119780000111
中的知识作为解码器的特征恢复路径。特征因果图图
Figure BDA0003908119780000112
和区域属性的顺序Π给出了解码器生成区域特征的顺序的信息,即解码器会按照特征因果图的顺序依次生成从根节点到叶节点的所有区域特征。每个区域特征
Figure BDA0003908119780000113
的条件分布是根据潜在变量Z及其父节点确定的。这样做的原因是希望潜在变量Z包含嵌入对象的完整配置文件。In the causal estimation enhanced decoder, the decoder recovers or reconstructs complete information including observed features and unobserved features based on latent variables Z (i.e., latent variables Z), such as p(X|Z) and p( Y|X,Z). Using Feature Causality Diagrams
Figure BDA0003908119780000111
The knowledge in is used as the feature recovery path of the decoder. Feature Causality Diagram
Figure BDA0003908119780000112
The order of Π and region attributes gives the information of the order in which the decoder generates region features, that is, the decoder will sequentially generate all region features from the root node to the leaf nodes in the order of the feature causality graph. characteristics of each area
Figure BDA0003908119780000113
The conditional distribution of is determined from the latent variable Z and its parents. The reason for this is that latent variable Z is expected to contain the full profile of the embedded object.

每个区域特征分布被重构的过程可以用下面的等式来概括。The process by which the feature distribution of each region is reconstructed can be summarized by the following equation.

Figure BDA0003908119780000114
Figure BDA0003908119780000114

Figure BDA0003908119780000115
Figure BDA0003908119780000115

其中

Figure BDA0003908119780000116
表示期望恢复的特征;
Figure BDA0003908119780000117
表示相应的高斯分布,μi和σi表示重构区域特征分布的均值与方差。in
Figure BDA0003908119780000116
Indicates the characteristics of the desired recovery;
Figure BDA0003908119780000117
Represents the corresponding Gaussian distribution, μ i and σ i represent the mean and variance of the feature distribution of the reconstructed region.

在初始的基于因果增强的变分自动编码器的训练过程中,CEVAE可以被视为将未观察到的区域特征Y显式建模为辅助潜在变量,并使用因果路径将特征之间的依赖关系添加到传统的vanillaVAE,以构建更有效的基于因果增强的变分自动编码器。与VAE类似,所述的因果增强变分自动编码器的积分优化目标是证据下界(ELBO),公式如下:During the training of the initial causal augmentation-based variational autoencoder, CEVAE can be viewed as explicitly modeling the unobserved regional feature Y as an auxiliary latent variable, and using causal paths to map the dependencies between features Added to traditional vanillaVAE to build more efficient causal augmentation based variational autoencoders. Similar to VAE, the integral optimization objective of the causally augmented variational autoencoder is the Evidence Lower Bound (ELBO), which is formulated as follows:

Figure BDA0003908119780000118
Figure BDA0003908119780000118

其中,q和p分别表示图3中编码器和解码器的条件分布;logq(Y=Y*|X)表示给定观测数据X下的Y*的似然;logq(Y=Y*|X,Z)给定观测数据X并得到的Z下的Y*的似然。Among them, q and p represent the conditional distributions of the encoder and decoder in Fig. 3 respectively; logq(Y=Y * |X) represents the likelihood of Y* under the given observation data X; logq(Y=Y * |X , Z) given the observed data X and obtained the likelihood of Y* under Z.

为了赋予辅助潜在变量Y物理约束,在ELBO中添加了两个额外项,如下:In order to give physical constraints to the auxiliary latent variable Y, two additional terms are added to ELBO, as follows:

Figure BDA0003908119780000121
Figure BDA0003908119780000121

其中,Y*表示缺失特征的观察标签;logq(Y=Y*|X)表示给定观测数据X下的Y*的似然;log q(Y=Y*|X,Z)给定观测数据X并得到的Z下的Y*的似然。这确保了Y的估计量和重建稳定学习,以避免在因果路径上顺序计算时积累错误和噪声。的优化目标是最大化ELBO。因此,用于优化因果增强变分自动编码器(即基于因果增强的变分自动编码器)的最终损失函数如下所示:Among them, Y * indicates the observation label of the missing feature; logq(Y=Y * |X) indicates the likelihood of Y* under the given observation data X; log q(Y=Y * |X,Z) given the observation data X and get the likelihood of Y* under Z. This ensures that the estimator of Y and the reconstruction are learned stably to avoid accumulating errors and noise while computing sequentially on the causal path. The optimization goal of is to maximize ELBO. Therefore, the final loss function for optimizing a causal augmented variational autoencoder (i.e., a causal augmented based variational autoencoder) looks like this:

Figure BDA0003908119780000122
Figure BDA0003908119780000122

其中,β是辅助损失的权重。where β is the weight of the auxiliary loss.

对于CEVAE的训练算法,随机抽取源区域中90%的区域特征构建训练数据集,其余10%的区域特征作为验证集用于调整超参数,并选择提前停止的时间以避免过度拟合。每个区域特征作为样本包含目标城市中观察到的特征集X作为输入和完整的特征集

Figure BDA0003908119780000123
作为重构输出。通过反向传播训练因果增强变异自动编码器,选择Adam作为优化器。For the training algorithm of CEVAE, 90% of the regional features in the source area are randomly extracted to construct a training data set, and the remaining 10% of the regional features are used as a validation set to adjust hyperparameters, and the time of early stopping is selected to avoid overfitting. Each regional feature as sample contains the feature set X observed in the target city as input and the full feature set
Figure BDA0003908119780000123
as refactored output. A causal augmented variational autoencoder is trained via backpropagation, and Adam is chosen as the optimizer.

步骤103:基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;其中,所述预测模型是预先基于所述源区域的数据构建的人口流动预测模型。所述知识蒸馏是一种在繁琐的模型(即基于数据丰富城市的预测模型)中提炼知识并将其压缩为目标模型(即发展中城市的人口流动预测模型)的方法,以便可以将其部署到数据稀疏的发展中城市的实际人口流动预测中。Step 103: Based on the transfer learning algorithm of knowledge distillation between the source area and the target area, transfer the knowledge of the prediction model to the target area, and based on the knowledge of the prediction model and the potential causal embedding vector Perform population flow prediction in the target area; wherein, the prediction model is a population flow prediction model constructed in advance based on the data of the source area. The knowledge distillation is a method of distilling knowledge in a cumbersome model (i.e., a predictive model based on a data-rich city) and compressing it into a target model (i.e., a population mobility prediction model for a developing city) so that it can be deployed into real population mobility forecasts for data-sparse developing cities.

在本发明实施例中,可基于所述预测模型的知识和所述潜在的因果嵌入向量获取所述目标区域中起点区域的特征和终点区域的特征,并添加起点和终点之间的距离信息,以进行所述目标区域的人口流动预测,即预测出目标区域中起点和终点之间的人流量。In the embodiment of the present invention, based on the knowledge of the prediction model and the potential causal embedding vector, the features of the start area and the end area in the target area can be obtained, and the distance information between the start point and the end point can be added, To predict the population flow in the target area, that is, to predict the flow of people between the start point and the end point in the target area.

具体的,在基于知识蒸馏的起点和终点之间人流量预测过程中,基于观察到的特征和学习到的嵌入向量的起点和终点之间人流量预测中使用的发展中城市的人口流动预测模型。为了突出的因果知识建模方法的有效性,可选择标准链接预测模式下的起点-终点人流量预测方法,结合原始区域的特征和目的区域的特征(即起点区域的特征和终点区域的特征),并添加距离信息来预测这对区域之间的流量。如图4所示,其为人口流动预测模型的预测过程。其中,Region Features表示区域特征,包括regionA、regionB、regionC、regionD、regionE;Urban topology表示城市的网络拓扑或区域的网络拓扑,GAT origin表示原始区域(即起点),GAT destination表示目的区域(即终点),Original Embedding表示原始区域嵌入向量,Destination Embedding表示目的区域嵌入向量;Distance matrix为一个距离矩阵,其是一个包含一组点两两之间距离的矩阵,destination regions表示目的区域地,original regions表示原始区域,Fetch by Index通过索引获取相应数据,prediction预测,iteration迭代,interation交互影响;Concatenate表示连接函数。Specifically, in the process of predicting the flow of people between the starting point and the destination based on knowledge distillation, the population flow prediction model of the developing city used in the prediction of the flow of people between the starting point and the ending point based on the observed features and the learned embedding vector . In order to highlight the effectiveness of the causal knowledge modeling method, the start-destination flow prediction method under the standard link prediction mode can be selected, combining the characteristics of the original area and the characteristics of the destination area (that is, the characteristics of the start area and the characteristics of the end area) , and add distance information to predict the flow between the pair of regions. As shown in Figure 4, it is the prediction process of the population flow prediction model. Among them, Region Features represent regional features, including regionA, regionB, regionC, regionD, regionE; Urban topology represents the network topology of a city or a region, GAT origin represents the original region (ie, the starting point), and GAT destination represents the destination area (ie, the destination ), Original Embedding represents the original region embedding vector, Destination Embedding represents the target region embedding vector; Distance matrix is a distance matrix, which is a matrix containing the distance between a group of points, destination regions represent the destination region, original regions represent In the original area, Fetch by Index obtains the corresponding data through the index, prediction prediction, iteration iteration, and interaction interaction; Concatenate represents the connection function.

具体过程如图4所示:显示的区域特征包括观察到的区域特征和学习到来自CEVAE模型的Z(或μzz)。然后在城市空间上构建图神经网络,以区域为节点,相邻区域连接边。由于基于空间连续性,相邻区域具有一定的相似性,使用图神经网络可以充分利用区域之间的相似性,使标签在城市空间上传播。使用GAT在城市空间的图网络上提取区域的空间特征,并使用原始特征和目的地特征来预测起点-终点流,另外,如图4所示,距离也是影响人口移动的关键因素,因此在预测中考虑了距离,并使用MSE(Microsoft SecurityEssentials)作为梯度下降的损失。针对发展中城市数据稀疏问题,在迁移预测模型时通过知识蒸馏方法来弥补目标城市(即目标区域)数据的稀疏性。The specific process is shown in Figure 4: the displayed regional features include the observed regional features and the learned Z (or μ z , σ z ) from the CEVAE model. Then a graph neural network is constructed on the urban space, with regions as nodes and adjacent regions connected as edges. Due to the spatial continuity, adjacent areas have a certain similarity, using graph neural network can make full use of the similarity between areas, so that labels can be propagated in urban space. Use GAT to extract the spatial features of the region on the graph network of urban space, and use the original features and destination features to predict the origin-destination flow. In addition, as shown in Figure 4, distance is also a key factor affecting population movement, so in predicting The distance is considered in , and MSE (Microsoft Security Essentials) is used as the loss of gradient descent. Aiming at the problem of data sparsity in developing cities, the knowledge distillation method is used to compensate for the data sparsity of the target city (that is, the target area) when migrating the prediction model.

为了更清楚地说明本发明实施例或现有技术中的技术方案,下面将对照图5说明本发明的具体实施方式。先在数据丰富的源区域采集所需要的数据,比如通过问卷调查,爬虫等方式。接着需要对采集到的数据进行清洗。然后会基于两种方式来辅助目标城市的预测。首先是通过强化学习因果发现的方法从采集到的数据中学到特征因果图,并在此基础上基于CEVAE得到观测特征恢复缺失特征的embedding,并把这个embedding迁移到目标城市。另一方面,在源区域训练特征预测OD-flow(起点到终点)的预测模型,并通过知识蒸馏的方式迁移到目标城市;最后基于得到的embedding和迁移到的预测模型,进行目标城市的OD-flow预测(即起点到终点的人口流动预测)。In order to more clearly illustrate the embodiment of the present invention or the technical solution in the prior art, the specific implementation manner of the present invention will be described below with reference to FIG. 5 . First collect the required data in the data-rich source area, such as through questionnaires, crawlers, etc. Next, the collected data needs to be cleaned. Then it will assist the prediction of the target city based on two methods. The first is to learn the characteristic causality map from the collected data through the method of reinforcement learning causality discovery, and on this basis, based on CEVAE, obtain the embedding of the observed features to restore the missing features, and transfer this embedding to the target city. On the other hand, the prediction model of feature prediction OD-flow (start point to end point) is trained in the source area, and migrated to the target city through knowledge distillation; finally, based on the obtained embedding and the transferred prediction model, the OD of the target city is performed -flow prediction (that is, the population flow prediction from the start point to the end point).

本发明实施例所述的数据稀疏区域的人口流动预测方法,通过基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识,并基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;然后,利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;进而基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;有效解决稀疏数据导致的预测困境,提高了针对数据稀疏区域的人口流动预测效率和精确度。The population flow prediction method in the data-sparse area described in the embodiment of the present invention obtains the corresponding regional causal knowledge from the data of the source area through the causal discovery model based on reinforcement learning, and automatically based on the regional causal knowledge and the initial variation Encoder to obtain a variational autoencoder based on causal enhancement; then, use the variational autoencoder based on causal enhancement to restore the missing features of the target region to obtain a potential causal embedding vector; then based on the source region A migration learning algorithm of knowledge distillation between the target area, transferring the knowledge of the prediction model to the target area, and performing population flow in the target area based on the knowledge of the prediction model and the potential causal embedding vector Forecasting: effectively solve the prediction dilemma caused by sparse data, and improve the efficiency and accuracy of population flow prediction for data-sparse areas.

与上述提供的一种数据稀疏区域的人口流动预测方法相对应,本发明还提供一种数据稀疏区域的人口流动预测装置。由于该装置的实施例相似于上述方法实施例,所以描述得比较简单,相关之处请参见上述方法实施例部分的说明即可,下面描述的数据稀疏区域的人口流动预测装置的实施例仅是示意性的。请参考图6所示,其为本发明实施例提供的一种数据稀疏区域的人口流动预测装置的结构示意图。Corresponding to the method for predicting population flow in a data-sparse area provided above, the present invention also provides a device for predicting population flow in a data-sparse area. Since the embodiment of the device is similar to the above-mentioned method embodiment, so the description is relatively simple, please refer to the description of the above-mentioned method embodiment for relevant information, and the embodiment of the population flow prediction device in the data-sparse area described below is only Schematic. Please refer to FIG. 6 , which is a schematic structural diagram of a population flow prediction device in a data-sparse area provided by an embodiment of the present invention.

本发明所述的数据稀疏区域的人口流动预测装置,具体包括如下部分:The population flow prediction device in the data-sparse area of the present invention specifically includes the following parts:

区域因果知识获取单元601,用于利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识;The regional causal knowledge acquisition unit 601 is used to obtain the corresponding regional causal knowledge from the data of the source region by using the causal discovery model based on reinforcement learning;

因果嵌入向量获取单元602,用于基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;其中,所述潜在的因果嵌入向量是观测特征以及所述缺失特征对应的表征向量;The causal embedding vector acquisition unit 602 is used to obtain a causal enhancement-based variational autoencoder based on the regional causal knowledge and the initial variational autoencoder; use the causal enhancement-based variational autoencoder to target the region The missing feature is recovered to obtain a potential causal embedding vector; wherein, the potential causal embedding vector is an observation feature and a representation vector corresponding to the missing feature;

人口流动预测单元603,用于基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;其中,所述预测模型是预先基于所述源区域的数据构建的人口流动预测模型。The population flow prediction unit 603 is configured to migrate the knowledge of the prediction model to the target area based on the migration learning algorithm of knowledge distillation between the source area and the target area, and based on the knowledge of the prediction model and the The potential causal embedding vector is used to predict the population flow of the target area; wherein, the prediction model is a population flow prediction model constructed in advance based on the data of the source area.

进一步的,所述区域因果知识获取单元,具体用于:Further, the regional causal knowledge acquisition unit is specifically used for:

获取所述源区域的数据;基于强化学习的区域因果知识构建策略确定基于区域属性特征排序的因果发现模型,利用所述因果发现模型对所述源区域的数据进行分析,以获得满足预设条件的区域属性特征顺序,并通过贝叶斯检验修剪得到包含区域属性特征之间关系的特征因果图;其中,所述特征因果图用于表示所述区域因果知识。Acquiring the data of the source area; determining the causal discovery model based on the ranking of regional attribute characteristics based on the regional causal knowledge construction strategy of reinforcement learning, and using the causal discovery model to analyze the data of the source area to obtain the The sequence of regional attribute features is obtained by Bayesian test pruning to obtain a feature causal graph containing the relationship between regional attribute features; wherein, the feature causal graph is used to represent the regional causal knowledge.

进一步的,所述因果嵌入向量获取单元,具体用于:Further, the causal embedding vector acquisition unit is specifically used for:

将所述特征因果图作为特征恢复路径,基于所述初始的变分自动编码器和所述特征恢复路径来学习包含区域属性特征的缺失信息,得到所述潜在的因果嵌入向量;所述潜在的因果嵌入向量为基于观测特征恢复的缺失特征的嵌入向量。Using the feature causal graph as a feature recovery path, learning missing information including regional attribute features based on the initial variational autoencoder and the feature recovery path, to obtain the potential causal embedding vector; the potential The causal embedding vector is the embedding vector of missing features based on observed feature recovery.

进一步的,所述因果嵌入向量获取单元,具体还用于:Further, the causal embedding vector acquisition unit is also specifically used for:

将所述目标区域的未观测到的缺失特征显式建模为辅助的潜在变量,并使用所述区域因果知识对应的因果路径将特征之间的关联关系添加到所述初始的变分自动编码器,以构建得到初始的基于因果增强的变分自动编码器;通过反向传播训练因果增强所述初始的基于因果增强的变分自动编码器,得到所述基于因果增强的变分自动编码器。Explicitly model the unobserved missing features of the target region as auxiliary latent variables, and use the causal path corresponding to the causal knowledge of the region to add the association relationship between features to the initial variational autoencoding To construct the initial variational autoencoder based on causal enhancement; through backpropagation training, causally enhance the initial variational autoencoder based on causal enhancement, and obtain the variational autoencoder based on causal enhancement .

进一步的,所述人口流动预测单元,具体用于:Further, the population flow prediction unit is specifically used for:

基于所述预测模型的知识和所述潜在的因果嵌入向量获取所述目标区域中起点区域的特征和终点区域的特征,并添加起点和终点之间的距离信息,来预测所述目标区域中起点和终点之间的人流量。Based on the knowledge of the predictive model and the potential causal embedding vector, obtain the features of the start area and the end area in the target area, and add the distance information between the start point and the end point to predict the start point in the target area traffic between the destination and the destination.

进一步的,所述区域因果知识为源区域中人员特征和区域属性特征之间的因果映射关系信息。Further, the regional causal knowledge is causal mapping relationship information between personnel characteristics and regional attribute characteristics in the source region.

进一步的,所述目标区域为数据稀疏区域。Further, the target area is a data-sparse area.

本发明实施例所述的数据稀疏区域的人口流动预测装置,通过基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识,并基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;然后,利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;进而基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;有效解决稀疏数据导致的预测困境,提高了针对数据稀疏区域的人口流动预测效率和精确度。The device for predicting population flow in a data-sparse area according to the embodiment of the present invention obtains the corresponding regional causal knowledge from the data of the source area through the causal discovery model based on reinforcement learning, and automatically based on the regional causal knowledge and the initial variation Encoder to obtain a variational autoencoder based on causal enhancement; then, use the variational autoencoder based on causal enhancement to restore the missing features of the target region to obtain a potential causal embedding vector; then based on the source region A migration learning algorithm of knowledge distillation between the target area, transferring the knowledge of the prediction model to the target area, and performing population flow in the target area based on the knowledge of the prediction model and the potential causal embedding vector Forecasting: effectively solve the prediction dilemma caused by sparse data, and improve the efficiency and accuracy of population flow prediction for data-sparse areas.

与上述提供的数据稀疏区域的人口流动预测方法相对应,本发明还提供一种电子设备。由于该电子设备的实施例相似于上述方法实施例,所以描述得比较简单,相关之处请参见上述方法实施例部分的说明即可,下面描述的电子设备仅是示意性的。如图7所示,其为本发明实施例公开的一种电子设备的实体结构示意图。该电子设备可以包括:处理器(processor)701、存储器(memory)702和通信总线703(即上述装置总线)以及查找引擎705,其中,处理器701,存储器702通过通信总线703完成相互间的通信,通过通信接口704与外部进行通信。处理器701可以调用存储器702中的逻辑指令,以执行数据稀疏区域的人口流动预测方法,该方法包括:利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识;基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;其中,所述潜在的因果嵌入向量是观测特征以及所述缺失特征对应的表征向量;基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;其中,所述预测模型是预先基于所述源区域的数据构建的人口流动预测模型。Corresponding to the method for predicting population flow in a data-sparse area provided above, the present invention also provides an electronic device. Since the embodiment of the electronic device is similar to the above-mentioned method embodiment, the description is relatively simple. For related details, please refer to the description of the above-mentioned method embodiment. The electronic device described below is only illustrative. As shown in FIG. 7 , it is a schematic diagram of a physical structure of an electronic device disclosed in an embodiment of the present invention. The electronic device may include: a processor (processor) 701, a memory (memory) 702, a communication bus 703 (ie, the above-mentioned device bus) and a search engine 705, wherein the processor 701 and the memory 702 complete mutual communication through the communication bus 703 , to communicate with the outside through the communication interface 704. The processor 701 can call the logic instructions in the memory 702 to execute the population flow prediction method in the data-sparse area. The method includes: using the causal discovery model based on reinforcement learning to obtain the corresponding regional causal knowledge from the data of the source area; The causal knowledge of the region and the initial variational autoencoder are used to obtain a variational autoencoder based on causal enhancement; the missing features of the target region are recovered by using the variational autoencoder based on causal enhancement to obtain a potential causal embedding Vector; wherein, the potential causal embedding vector is the representation vector corresponding to the observation feature and the missing feature; based on the migration learning algorithm of knowledge distillation between the source area and the target area, the knowledge of the prediction model is transferred to the target area, and perform population flow prediction in the target area based on the knowledge of the prediction model and the potential causal embedding vector; wherein, the prediction model is a population flow constructed in advance based on the data of the source area predictive model.

此外,上述的存储器702中的逻辑指令可以通过软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括:存储芯片、U盘、移动硬盘、只读存储器(ROM,Read-Only Memory)、随机存取存储器(RAM,Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。In addition, the above-mentioned logic instructions in the memory 702 may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include: memory chips, U disks, mobile hard disks, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks or optical disks, etc., which can store programs. The medium of the code.

另一方面,本发明实施例还提供一种计算机程序产品,所述计算机程序产品包括存储在处理器可读存储介质上的计算机程序,所述计算机程序包括程序指令,当所述程序指令被计算机执行时,计算机能够执行上述各方法实施例所提供的数据稀疏区域的人口流动预测方法。该方法包括:利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识;基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;其中,所述潜在的因果嵌入向量是观测特征以及所述缺失特征对应的表征向量;基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;其中,所述预测模型是预先基于所述源区域的数据构建的人口流动预测模型。On the other hand, an embodiment of the present invention also provides a computer program product, the computer program product includes a computer program stored on a processor-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by the computer During execution, the computer can execute the method for predicting population flow in data-sparse regions provided by the above method embodiments. The method includes: using a causal discovery model based on reinforcement learning to obtain the corresponding regional causal knowledge from the data of the source region; based on the regional causal knowledge and the initial variational autoencoder, a variational autoencoder based on causal enhancement is obtained. ; Utilizing the variational autoencoder based on causal enhancement to restore the missing features of the target region to obtain a potential causal embedding vector; wherein the potential causal embedding vector is an observation feature and a representation vector corresponding to the missing feature ; Based on a migration learning algorithm of knowledge distillation between the source area and the target area, the knowledge of the prediction model is migrated to the target area, and based on the knowledge of the prediction model and the potential causal embedding vector. The population flow prediction of the target area; wherein, the prediction model is a population flow prediction model constructed in advance based on the data of the source area.

又一方面,本发明实施例还提供一种处理器可读存储介质,所述处理器可读存储介质上存储有计算机程序,该计算机程序被处理器执行时实现以执行上述各实施例提供的数据稀疏区域的人口流动预测方法。该方法包括:利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识;基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;其中,所述潜在的因果嵌入向量是观测特征以及所述缺失特征对应的表征向量;基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;其中,所述预测模型是预先基于所述源区域的数据构建的人口流动预测模型。In yet another aspect, an embodiment of the present invention further provides a processor-readable storage medium, where a computer program is stored on the processor-readable storage medium, and the computer program is implemented when executed by a processor to perform the functions provided by the above-mentioned embodiments. Population mobility prediction methods for data-sparse regions. The method includes: using a causal discovery model based on reinforcement learning to obtain the corresponding regional causal knowledge from the data of the source region; based on the regional causal knowledge and the initial variational autoencoder, a variational autoencoder based on causal enhancement is obtained. ; Utilizing the variational autoencoder based on causal enhancement to restore the missing features of the target region to obtain a potential causal embedding vector; wherein the potential causal embedding vector is an observation feature and a representation vector corresponding to the missing feature ; Based on a migration learning algorithm of knowledge distillation between the source area and the target area, the knowledge of the prediction model is migrated to the target area, and based on the knowledge of the prediction model and the potential causal embedding vector. The population flow prediction of the target area; wherein, the prediction model is a population flow prediction model constructed in advance based on the data of the source area.

所述处理器可读存储介质可以是处理器能够存取的任何可用介质或数据存储设备,包括但不限于磁性存储器(例如软盘、硬盘、磁带、磁光盘(MO)等)、光学存储器(例如CD、DVD、BD、HVD等)、以及半导体存储器(例如ROM、EPROM、EEPROM、非易失性存储器(NANDFLASH)、固态硬盘(SSD))等。The processor-readable storage medium can be any available medium or data storage device that can be accessed by a processor, including but not limited to magnetic storage (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.), optical storage (e.g., CD, DVD, BD, HVD, etc.), and semiconductor memory (such as ROM, EPROM, EEPROM, non-volatile memory (NANDFLASH), solid-state disk (SSD)), etc.

以上所描述的装置实施例仅仅是示意性的,其中所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。本领域普通技术人员在不付出创造性的劳动的情况下,即可以理解并实施。The device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in One place, or it can be distributed to multiple network elements. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. It can be understood and implemented by those skilled in the art without any creative effort.

通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到各实施方式可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件。基于这样的理解,上述技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品可以存储在计算机可读存储介质中,如ROM/RAM、磁碟、光盘等,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行各个实施例或者实施例的某些部分所述的方法。Through the above description of the implementations, those skilled in the art can clearly understand that each implementation can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic discs, optical discs, etc., including several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) execute the methods described in various embodiments or some parts of the embodiments.

最后应说明的是:以上实施例仅用以说明本发明的技术方案,而非对其限制;尽管参照前述实施例对本发明进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims (10)

1.一种数据稀疏区域的人口流动预测方法,其特征在于,包括:1. A population flow prediction method in a data-sparse area, characterized in that it comprises: 利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识;Use a causal discovery model based on reinforcement learning to obtain the corresponding regional causal knowledge from the data of the source region; 基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;其中,所述潜在的因果嵌入向量是观测特征以及所述缺失特征对应的表征向量;Based on the causal knowledge of the region and the initial variational autoencoder, a variational autoencoder based on causal enhancement is obtained; the missing features of the target region are recovered by using the variational autoencoder based on causal enhancement to obtain potential A causal embedding vector; wherein, the potential causal embedding vector is an observation feature and a representation vector corresponding to the missing feature; 基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;其中,所述预测模型是预先基于所述源区域的数据构建的人口流动预测模型。Based on the transfer learning algorithm of knowledge distillation between the source area and the target area, the knowledge of the prediction model is transferred to the target area, and the knowledge of the prediction model and the potential causal embedding vector are used to perform the Population flow prediction in the target area; wherein, the prediction model is a population flow prediction model constructed in advance based on the data of the source area. 2.根据权利要求1所述的数据稀疏区域的人口流动预测方法,其特征在于,所述利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识,具体包括:2. The method for predicting population flow in a data-sparse area according to claim 1, wherein said utilization of a causal discovery model based on reinforcement learning obtains corresponding regional causal knowledge from data in the source area, specifically comprising: 获取所述源区域的数据;基于强化学习的区域因果知识构建策略确定基于区域属性特征排序的因果发现模型,利用所述因果发现模型对所述源区域的数据进行分析,以获得满足预设条件的区域属性特征顺序,并通过贝叶斯检验修剪得到包含区域属性特征之间关系的特征因果图;其中,所述特征因果图用于表示所述区域因果知识。Acquiring the data of the source area; determining the causal discovery model based on the ranking of regional attribute characteristics based on the regional causal knowledge construction strategy of reinforcement learning, and using the causal discovery model to analyze the data of the source area to obtain the The sequence of regional attribute features is obtained by Bayesian test pruning to obtain a feature causal graph containing the relationship between regional attribute features; wherein, the feature causal graph is used to represent the regional causal knowledge. 3.根据权利要求2所述的数据稀疏区域的人口流动预测方法,其特征在于,利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量,具体包括:3. The method for predicting population flow in data-sparse regions according to claim 2, characterized in that, using the causally enhanced variational autoencoder to restore the missing features of the target region to obtain potential causal embedding vectors, Specifically include: 将所述特征因果图作为特征恢复路径,基于所述初始的变分自动编码器和所述特征恢复路径来学习包含区域属性特征的缺失信息,得到所述潜在的因果嵌入向量。The feature causal graph is used as a feature recovery path, and the missing information including region attribute features is learned based on the initial variational autoencoder and the feature recovery path to obtain the potential causal embedding vector. 4.根据权利要求1所述的数据稀疏区域的人口流动预测方法,其特征在于,基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器,具体包括:4. The method for predicting population flow in data-sparse regions according to claim 1, characterized in that, based on the regional causal knowledge and initial variational autoencoder, obtain a variational autoencoder based on causal enhancement, specifically comprising : 将所述目标区域的未观测到的缺失特征显式建模为辅助的潜在变量,并使用所述区域因果知识对应的因果路径将特征之间的关联关系添加到所述初始的变分自动编码器,以构建得到初始的基于因果增强的变分自动编码器;通过反向传播训练因果增强所述初始的基于因果增强的变分自动编码器,得到所述基于因果增强的变分自动编码器。Explicitly model the unobserved missing features of the target region as auxiliary latent variables, and use the causal path corresponding to the causal knowledge of the region to add the association relationship between features to the initial variational autoencoding To construct the initial variational autoencoder based on causal enhancement; through backpropagation training, causally enhance the initial variational autoencoder based on causal enhancement, and obtain the variational autoencoder based on causal enhancement . 5.根据权利要求1所述的数据稀疏区域的人口流动预测方法,其特征在于,基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测,具体包括:5. The population flow prediction method in the data-sparse area according to claim 1, characterized in that, based on the knowledge of the prediction model and the potential causal embedding vector, the population flow prediction of the target area is carried out, specifically comprising: 基于所述预测模型的知识和所述潜在的因果嵌入向量获取所述目标区域中起点区域的特征和终点区域的特征,并添加起点和终点之间的距离信息,来预测所述目标区域中起点和终点之间的人流量。Based on the knowledge of the predictive model and the potential causal embedding vector, obtain the features of the start area and the end area in the target area, and add the distance information between the start point and the end point to predict the start point in the target area traffic between the destination and the destination. 6.根据权利要求1所述的数据稀疏区域的人口流动预测方法,其特征在于,所述区域因果知识为源区域中人员特征和区域属性特征之间的因果映射关系信息。6. The method for predicting population flow in a data-sparse area according to claim 1, wherein the regional causal knowledge is causal mapping relationship information between personnel characteristics and regional attribute characteristics in the source region. 7.根据权利要求1所述的数据稀疏区域的人口流动预测方法,其特征在于,所述目标区域为数据稀疏区域。7. The method for predicting population flow in a data-sparse area according to claim 1, wherein the target area is a data-sparse area. 8.一种数据稀疏区域的人口流动预测装置,其特征在于,包括:8. A population flow prediction device in a data-sparse area, characterized in that it comprises: 区域因果知识获取单元,用于利用基于强化学习的因果发现模型从源区域的数据中获取相应的区域因果知识;The regional causal knowledge acquisition unit is used to obtain the corresponding regional causal knowledge from the data of the source region by using the causal discovery model based on reinforcement learning; 因果嵌入向量获取单元,用于基于所述区域因果知识和初始的变分自动编码器,得到基于因果增强的变分自动编码器;利用所述基于因果增强的变分自动编码器对目标区域的缺失特征进行恢复,获得潜在的因果嵌入向量;其中,所述潜在的因果嵌入向量是观测特征以及所述缺失特征对应的表征向量;The causal embedding vector acquisition unit is used to obtain a causal-enhanced variational autoencoder based on the regional causal knowledge and an initial variational autoencoder; using the causal-enhanced variational autoencoder for the target region The missing feature is restored to obtain a potential causal embedding vector; wherein the potential causal embedding vector is an observation feature and a representation vector corresponding to the missing feature; 人口流动预测单元,用于基于所述源区域与所述目标区域之间知识蒸馏的迁移学习算法,将预测模型的知识迁移到所述目标区域,并基于所述预测模型的知识和所述潜在的因果嵌入向量进行所述目标区域的人口流动预测;其中,所述预测模型是预先基于所述源区域的数据构建的人口流动预测模型。The population flow prediction unit is used to transfer the knowledge of the prediction model to the target area based on the migration learning algorithm of knowledge distillation between the source area and the target area, and based on the knowledge of the prediction model and the potential The causal embedding vector is used to predict the population flow of the target area; wherein, the prediction model is a population flow prediction model constructed in advance based on the data of the source area. 9.一种电子设备,包括存储器、处理器以及存储在存储器上并可在处理器上运行的计算机程序,其特征在于,所述处理器执行所述计算机程序时实现如权利要求1至7任意一项所述的数据稀疏区域的人口流动预测方法的步骤。9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, characterized in that, when the processor executes the computer program, any of claims 1 to 7 is realized. A method for predicting population flow in a data-sparse region. 10.一种处理器可读存储介质,所述处理器可读存储介质上存储有计算机程序,其特征在于,该计算机程序被处理器执行时实现如权利要求1至7任意一项所述的数据稀疏区域的人口流动预测方法的步骤。10. A processor-readable storage medium, with a computer program stored on the processor-readable storage medium, characterized in that, when the computer program is executed by a processor, the computer program according to any one of claims 1 to 7 is implemented. Steps in a population mobility prediction method for data-sparse regions.
CN202211313718.7A 2022-10-25 2022-10-25 Population flow prediction method and device for data sparse region Active CN115759350B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211313718.7A CN115759350B (en) 2022-10-25 2022-10-25 Population flow prediction method and device for data sparse region

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211313718.7A CN115759350B (en) 2022-10-25 2022-10-25 Population flow prediction method and device for data sparse region

Publications (2)

Publication Number Publication Date
CN115759350A true CN115759350A (en) 2023-03-07
CN115759350B CN115759350B (en) 2024-07-12

Family

ID=85353167

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211313718.7A Active CN115759350B (en) 2022-10-25 2022-10-25 Population flow prediction method and device for data sparse region

Country Status (1)

Country Link
CN (1) CN115759350B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118643949A (en) * 2024-08-15 2024-09-13 北京航空航天大学 Urban spatiotemporal data prediction method based on time-guided causal structure learning

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160026942A1 (en) * 2014-07-24 2016-01-28 Optum, Inc. System and method for identifying relationships in community healthcare measures
CN111612206A (en) * 2020-03-30 2020-09-01 清华大学 A method and system for predicting pedestrian flow in blocks based on spatiotemporal graph convolutional neural network
US20210174216A1 (en) * 2019-12-04 2021-06-10 International Business Machines Corporation Signaling concept drift during knowledge base population
CN113052635A (en) * 2021-03-30 2021-06-29 北京明略昭辉科技有限公司 Population attribute label prediction method, system, computer device and storage medium
WO2021174876A1 (en) * 2020-09-18 2021-09-10 平安科技(深圳)有限公司 Smart decision-based population movement prediction method, apparatus, and computer device
CN113610309A (en) * 2021-08-13 2021-11-05 清华大学 Method and device for site selection of fire station based on big data and artificial intelligence
CN113642807A (en) * 2021-09-01 2021-11-12 智慧足迹数据科技有限公司 Population mobility prediction method and related device
US20210366603A1 (en) * 2019-09-25 2021-11-25 Brilliance Center B.V. Methods for anonymously tracking and/or analysing health in a population of subjects
CN114239983A (en) * 2021-12-22 2022-03-25 广东电网有限责任公司 Target area population flow prediction method and related device

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160026942A1 (en) * 2014-07-24 2016-01-28 Optum, Inc. System and method for identifying relationships in community healthcare measures
US20210366603A1 (en) * 2019-09-25 2021-11-25 Brilliance Center B.V. Methods for anonymously tracking and/or analysing health in a population of subjects
US20210174216A1 (en) * 2019-12-04 2021-06-10 International Business Machines Corporation Signaling concept drift during knowledge base population
CN111612206A (en) * 2020-03-30 2020-09-01 清华大学 A method and system for predicting pedestrian flow in blocks based on spatiotemporal graph convolutional neural network
WO2021174876A1 (en) * 2020-09-18 2021-09-10 平安科技(深圳)有限公司 Smart decision-based population movement prediction method, apparatus, and computer device
CN113052635A (en) * 2021-03-30 2021-06-29 北京明略昭辉科技有限公司 Population attribute label prediction method, system, computer device and storage medium
CN113610309A (en) * 2021-08-13 2021-11-05 清华大学 Method and device for site selection of fire station based on big data and artificial intelligence
CN113642807A (en) * 2021-09-01 2021-11-12 智慧足迹数据科技有限公司 Population mobility prediction method and related device
CN114239983A (en) * 2021-12-22 2022-03-25 广东电网有限责任公司 Target area population flow prediction method and related device

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
WANG, MINJIE: "Human mobility prediction from region functions with taxi trajectories", 《PLOS ONE》, vol. 12, no. 11, 30 November 2017 (2017-11-30), pages 1 - 15 *
乐志强: "社会阶层认知和教育水平对流动人口义务教育公共服务满意度的影响", 《上海教育科研》, no. 12, 31 December 2016 (2016-12-31), pages 18 - 22 *
张燕;: "城市群的形成机理研究", 城市与环境研究, no. 01, 20 September 2014 (2014-09-20), pages 22 - 25 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118643949A (en) * 2024-08-15 2024-09-13 北京航空航天大学 Urban spatiotemporal data prediction method based on time-guided causal structure learning
CN118643949B (en) * 2024-08-15 2024-11-01 北京航空航天大学 Urban spatiotemporal data prediction method based on time-guided causal structure learning

Also Published As

Publication number Publication date
CN115759350B (en) 2024-07-12

Similar Documents

Publication Publication Date Title
US11521221B2 (en) Predictive modeling with entity representations computed from neural network models simultaneously trained on multiple tasks
US11907675B2 (en) Generating training datasets for training neural networks
CN103201754B (en) Data handling equipment and data processing method
CN113988464B (en) Network link attribute relation prediction method and device based on graph neural network
CN114299723A (en) Traffic flow prediction method
CN115376318B (en) Traffic data compensation method based on multi-attribute fusion neural network
CN117116048A (en) Knowledge-driven traffic prediction method based on knowledge representation model and graph neural network
CN117688453B (en) A traffic flow prediction method based on spatiotemporal embedded attention network
CN116596151B (en) Traffic flow prediction method and computing equipment based on spatiotemporal graph attention
CN114360239A (en) A traffic prediction method and system for multi-layer spatiotemporal traffic knowledge graph reconstruction
CN111369299A (en) Method, device and equipment for identification and computer readable storage medium
CN115860179B (en) Track prediction method, track prediction device, track prediction apparatus, track prediction storage medium, and track prediction program product
WO2025092993A1 (en) Electrical load prediction method and device based on spatial-temporal correlation
CN111000492B (en) Intelligent sweeper behavior decision method based on knowledge graph and intelligent sweeper
Yuan et al. Route travel time estimation on a road network revisited: Heterogeneity, proximity, periodicity and dynamicity
CN115759350B (en) Population flow prediction method and device for data sparse region
CN117173885A (en) Traffic state prediction method and system based on dynamic graph feature learning
CN114724630B (en) Deep learning method for predicting post-translational modification site of protein
Goenawan ASTM: Autonomous Smart Traffic Management System Using Artificial Intelligence CNN and LSTM
CN115860798A (en) Urban taxi demand prediction system and method
CN115423162A (en) A traffic flow prediction method, device, electronic equipment and storage medium
CN114911969A (en) A recommendation strategy optimization method and system based on user behavior model
CN116955335B (en) Address data management method and system based on big data model algorithm
CN118410915A (en) Method, device, equipment and medium for predicting OD (optical density) pair area demand of inter-city carpool based on STZINB-GCN (graphics communication network)
Poornima et al. Prediction of water consumption using machine learning algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant