Return-to-country information prediction method and device, storage medium and computer equipment

Abstract

The invention discloses a return information prediction method and device, a storage medium and computer equipment, relates to the technical field of information processing, and mainly aims to improve the prediction precision of return information of a user so as to improve the planning rationality of power grid capacity. The method comprises the following steps: in response to a return information prediction signal of a target object in a target holiday, obtaining object feature data, power consumption feature data and historical return data of the target object; performing feature cross enhancement processing on the object feature data, the power consumption feature data and the historical return data to obtain a return cross enhancement vector; and inputting the return cross enhancement vector into a preset return information prediction model for return information prediction to obtain a potential return result of the target object in the target holiday. The method is suitable for the scene of predicting the return information of the user.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to a method, apparatus, storage medium, and computer equipment for predicting information on people returning home. Background Technology

[0002] Users returning to their hometowns often cause significant changes in electricity load during specific time periods and in different regions. By accurately predicting this return-to-hometown information, power grid companies can more precisely forecast electricity load in different regions and at different times, thereby rationally planning grid capacity and ensuring the stability of users' electricity supply.

[0003] Currently, the prediction of whether a user will return home for a future holiday is usually based solely on their daily electricity consumption during past holidays. However, relying solely on historical daily electricity consumption as the basis for determining whether or not a user will return home provides limited information, resulting in low accuracy in predicting user return home status. Summary of the Invention

[0004] This invention provides a method, apparatus, storage medium, and computer equipment for predicting return-home information, which mainly improves the accuracy of predicting users' return-home information, thereby enhancing the rationality of power grid capacity planning.

[0005] According to a first aspect of the present invention, a method for predicting return-home information is provided, comprising: In response to the prediction signal of the target object's return home information during the target holiday, the target object's object characteristic data, electricity consumption characteristic data, and historical return home data are acquired; The object feature data, the electricity consumption feature data, and the historical return-home data are subjected to feature cross-enhancement processing to obtain the return-home cross-enhancement vector; The cross-enhancement vector for returning home is input into a preset return home information prediction model to predict return home information, thereby obtaining the potential return home results of the target object during the target holiday.

[0006] Optionally, before inputting the return-home cross-enhancement vector into a preset return-home information prediction model to predict return-home information and obtain the return-home result of the target object, the method further includes: Construct a pre-defined initial return-home information prediction model; Obtain a sample dataset, wherein the sample dataset includes sample object feature data, sample electricity consumption feature data, and sample historical return home data of sample objects with holiday return home information tags; The sample dataset is divided into a training set and a test set. The training set is used to train the preset initial return home information prediction model, and the test set is used to test the trained preset initial return home information prediction model. Finally, the trained preset initial return home information prediction model that meets the test conditions is taken as the preset return home information prediction model.

[0007] Optionally, obtaining the sample dataset includes: Use a preset population inflow and outflow model to identify users who have experienced population outflow during a preset period and whose homes are currently vacant; Based on historical data from the same period, identify users who return home in a regular pattern during the same period in history, wherein the historical data refers to the activity range of users during historical holidays of the same type as the target holiday; The users that intersect with the users to be identified and the users returning home are extracted as the sample objects. The sample object feature data, sample electricity consumption feature data, and sample historical return home data of the sample objects are obtained. The sample dataset is composed of the sample objects and their corresponding sample object feature data, sample electricity consumption feature data, and sample historical return home data.

[0008] Optionally, the step of performing feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return-home data to obtain the return-home cross-enhancement vector includes: Determine the object feature vector corresponding to the object feature data, the electricity consumption feature vector corresponding to the electricity consumption feature data, and the return home feature vector corresponding to the historical return home data, respectively. The object feature vector, the electricity consumption feature vector, and the return-home feature vector are subjected to feature-level cross-enhancement processing to obtain the feature cross-enhancement vector; Element-level cross-enhancement processing is performed on the object feature vector, the electricity consumption feature vector, and the return-home feature vector to obtain the element-level cross-enhancement vector; The object feature vector, the electricity consumption feature vector, and the return-home feature vector are subjected to low-order cross-enhancement processing to obtain a low-order cross-enhancement vector. The feature cross-enhancement vector, the element cross-enhancement vector, and the low-order cross-enhancement vector are subjected to enhancement transformation to obtain the homecoming cross-enhancement vector.

[0009] Optionally, before performing feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return-home data to obtain the return-home cross-enhancement vector, the method further includes: Each of the object feature data, the electricity consumption feature data, and the historical return-home data can be used as a target feature data. Take any data point in the target feature data as a target point, determine the initial neighborhood corresponding to the target point, and form a covariance matrix from the neighborhood points in the initial neighborhood. The covariance matrix is ​​decomposed into eigenvalues ​​and corresponding eigenvectors. The eigenvector corresponding to the smallest eigenvalue is used as the normal vector. The distance difference between the maximum and minimum points in the direction of the normal vector is determined. Based on the distance difference, the neighborhood angle of the initial neighborhood is determined. Based on the neighborhood angle, the initial neighborhood is adjusted to obtain the ultimate neighborhood corresponding to the target point. The mean and standard deviation of all points within the ultimate neighborhood are determined. Based on the mean and standard deviation, the absolute data evaluation threshold and the relative data evaluation threshold are determined respectively. Based on the absolute evaluation threshold and the relative evaluation threshold of the data, noise judgment is performed on the target feature data represented by the target point. Based on the noise judgment result, noise points are identified in the target feature data and the noise points in the target feature data are removed to obtain the denoised target feature data. The step of performing feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return-home data to obtain the return-home cross-enhancement vector includes: The denoised object feature data, the denoised electricity consumption feature data, and the denoised historical return-home data are subjected to feature cross-enhancement processing to obtain the return-home cross-enhancement vector.

[0010] Optionally, the step of inputting the return-home cross-enhancement vector into a preset return-home information prediction model to predict return-home information and obtain the potential return-home results of the target object during the target holiday includes: Multiple groups of people returning to their hometowns are identified, and the centroid vector corresponding to each group is determined. Based on the similarity between the cross-enhancing vector for returning home and each centroid vector, the target returning home information group to which the target object belongs is determined in each returning home information group; Determine the group return information of the target group, and input the group return information and the return cross-enhancement vector into a preset return information prediction model to predict the return information, thereby obtaining the potential return results of the target object during the target holiday.

[0011] Optionally, the step of inputting the return-home cross-enhancement vector into a preset return-home information prediction model to predict return-home information and obtain the potential return-home results of the target object during the target holiday includes: Obtain the network behavior data and communication data of the target object; The network behavior data and the communication data are fused to obtain a fused feature vector; The fused feature vector and the homecoming cross-enhancement vector are horizontally concatenated to obtain a concatenated feature vector. The concatenated feature vector is then input into the preset homecoming information prediction model to predict homecoming information, thereby obtaining the potential homecoming result of the target object during the target holiday.

[0012] According to a second aspect of the present invention, a device for predicting return-home information is provided, comprising: The acquisition unit is used to acquire the target object's object feature data, electricity consumption feature data, and historical return home data in response to the target object's return home information prediction signal during the target holiday; The cross-enhancement unit is used to perform feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return-home data to obtain the return-home cross-enhancement vector. The homecoming prediction unit is used to input the homecoming cross-enhancement vector into a preset homecoming information prediction model to predict homecoming information and obtain the potential homecoming results of the target object during the target holiday.

[0013] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described homecoming information prediction device.

[0014] According to a fourth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-mentioned homecoming information prediction device.

[0015] According to the present invention, a method, apparatus, storage medium, and computer device for predicting homecoming information are provided. Compared with the current method of judging a user's homecoming situation in future holidays based solely on the user's daily electricity consumption during past holidays, the present invention obtains the target object's object feature data, electricity consumption feature data, and historical homecoming data in response to the target object's homecoming information prediction signal during the target holiday. Then, it performs feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical homecoming data to obtain a homecoming cross-enhancement vector. Finally, it inputs the homecoming cross-enhancement vector into a preset homecoming information prediction model to predict homecoming information and obtain the potential homecoming result of the target object during the target holiday. Therefore, by comprehensively analyzing multi-dimensional information such as object characteristic data, electricity consumption characteristic data, and historical return home data, the accuracy of predicting the return home situation of the target object can be improved. By performing feature cross-enhancement processing on object characteristic data, electricity consumption characteristic data, and historical return home data, more latent features can be extracted, and the relationships between data can be fully utilized to make the subsequent prediction results more accurate. By using a pre-set return home information prediction model to predict the return home situation of the target object, the prediction accuracy and efficiency of return home information can be further improved. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart of a method for predicting return-to-hometown information provided by an embodiment of the present invention is shown; Figure 2 This invention provides a flowchart of another method for predicting return-home information. Figure 3 This diagram illustrates the structure of a homecoming information prediction device provided in an embodiment of the present invention. Figure 4 This invention provides a schematic diagram of another homecoming information prediction device according to an embodiment of the invention. Figure 5 A schematic diagram of the physical structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0017] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0018] Currently, the method of judging users' return home during future holidays based solely on their daily electricity consumption during past holidays is limited by information and results in low accuracy in predicting users' return home.

[0019] To address the aforementioned problems, embodiments of the present invention provide a method for predicting return-to-hometown information, such as... Figure 1 As shown, the method includes: 101. In response to the target object's homecoming information prediction signal during the target holiday, acquire the target object's object characteristic data, electricity consumption characteristic data, and historical homecoming data.

[0020] The target object can be any electricity user; the target holiday; the target characteristic data includes, but is not limited to, residential electricity users, urban or rural attributes of the object (such as city or rural), age, occupation, family members, etc., as displayed in the object file; the electricity consumption characteristic data includes, but is not limited to, the target object's electricity consumption number, due date, energy consumption, amount due, amount received, settlement status, outstanding balance, energy consumption category, etc., as well as the target object's electricity consumption, electricity consumption change rate, daily electricity consumption data, etc., within a preset time period, for the past 12 months; historical return home data refers to the target object's historical return home behavior data. This embodiment of the invention determines the target object's return home situation by comprehensively analyzing multi-dimensional information such as target characteristic data, electricity consumption characteristic data, and historical return home data, which can improve the prediction accuracy of return home information.

[0021] It should be noted that the target object's characteristic data, electricity consumption characteristic data, and historical return home data obtained in this embodiment are not the target object's personal privacy data, but rather data related to the target object's electricity consumption that can be obtained from the power grid management platform (not personal privacy data).

[0022] 102. Perform feature cross-enhancement processing on object feature data, electricity consumption feature data, and historical return-home data to obtain the return-home cross-enhancement vector.

[0023] In this embodiment of the invention, to fully utilize various data and extract latent features, it is necessary to perform feature cross-enhancement processing on object feature data, electricity consumption feature data, and historical return-home data, thereby improving the prediction accuracy of subsequent return-home information. This embodiment of the invention, through cross-enhancement processing of object feature data, electricity consumption feature data, and historical return-home data, can automatically or explicitly combine different features to generate new feature combinations. These combined features may contain complex nonlinear relationships between the original features, enabling the model to capture more refined and richer information from the data. This means it can fully utilize the relationships between various data, extract more latent features, make more efficient use of the data, and obtain more accurate prediction results for subsequent return-home information. This, in turn, improves the rationality of power grid power planning, ensures the stable operation of the power grid, and enhances the user's electricity experience.

[0024] 103. Input the cross-enhancement vector of returning home into the preset returning home information prediction model to predict the returning home information and obtain the potential returning home results of the target object during the target holiday.

[0025] Among them, the potential return-to-hometown outcome can be the probability of the target object returning to their hometown during the target holiday.

[0026] In this embodiment of the invention, after determining the homecoming cross-enhancement vector, the vector is directly input into a preset homecoming information prediction model. This model can then directly output the potential homecoming results of the target object during the target holiday. This embodiment of the invention uses a model to predict the homecoming information of the object, thereby improving the prediction efficiency and accuracy of homecoming information.

[0027] Furthermore, after predicting the potential return-to-hometown information of all individuals in a certain region or city, the electricity load of that region can be predicted based on this potential return-to-hometown information, providing a reliable basis for power grid dispatch, avoiding power shortages or surpluses due to load prediction deviations, improving the stability of residents' electricity use, and thus enhancing residents' electricity experience.

[0028] According to the present invention, a method for predicting homecoming information, compared with the current method of judging a user's homecoming situation in future holidays solely based on the user's daily electricity consumption during past holidays, this invention, in response to a homecoming information prediction signal for a target object during a target holiday, acquires the target object's object feature data, electricity consumption feature data, and historical homecoming data; then, it performs feature cross-enhancement processing on the object feature data, electricity consumption feature data, and historical homecoming data to obtain a homecoming cross-enhancement vector; finally, it inputs the homecoming cross-enhancement vector into a preset homecoming information prediction model to predict homecoming information, thereby obtaining the potential homecoming result of the target object during the target holiday. Thus, by comprehensively analyzing multi-dimensional information such as object feature data, electricity consumption feature data, and historical homecoming data to determine the target object's homecoming situation, the prediction accuracy of the object's homecoming situation can be improved; by performing feature cross-enhancement processing on the object feature data, electricity consumption feature data, and historical homecoming data, more latent features can be extracted, and the relationships between data can be fully utilized, making the subsequent prediction results more accurate; by using a preset homecoming information prediction model to predict the object's homecoming situation, the prediction accuracy and efficiency of homecoming information can be further improved.

[0029] Furthermore, to better illustrate the process of predicting return-home information described above, as a refinement and extension of the above embodiments, this invention provides another method for predicting return-home information, such as... Figure 2 As shown, the method includes: 201. In response to the target object's homecoming information prediction signal during the target holiday, acquire the target object's object characteristic data, electricity consumption characteristic data, and historical homecoming data.

[0030] Specifically, the target object's characteristic data, electricity consumption characteristic data, and historical return-to-hometown data are obtained from the power grid database.

[0031] 202. Perform feature cross-enhancement processing on object feature data, electricity consumption feature data, and historical return-home data to obtain the return-home cross-enhancement vector.

[0032] In this embodiment of the invention, to improve data quality, noise reduction processing is required before performing cross-enhancement processing. Based on this, the method includes: taking any one of the object feature data, the electricity consumption feature data, and the historical return-home data as a target feature data, taking any one of the target feature data as a target point, determining the initial neighborhood corresponding to the target point, and constructing a covariance matrix from the neighborhood points within the initial neighborhood; performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors corresponding to the eigenvalues, taking the eigenvector corresponding to the minimum eigenvalue as the normal vector, and determining the maximum and minimum points of the normal vector direction. The distance difference between small points is used to determine the neighborhood angle of the initial neighborhood. Based on the neighborhood angle, the initial neighborhood is adjusted to obtain the ultimate neighborhood corresponding to the target point. The mean and standard deviation of all points within the ultimate neighborhood are determined. Based on the mean and standard deviation, the absolute evaluation threshold and the relative evaluation threshold are determined respectively. Based on the absolute evaluation threshold and the relative evaluation threshold, noise judgment is performed on the target feature data represented by the target point. Based on the noise judgment result, noise points are identified in the target feature data and removed from the target feature data to obtain the denoised target feature data.

[0033] Specifically, with Using the radius as the radius and the target point as the center, an initial neighborhood is determined. Principal component analysis is used to calculate the covariance matrix of each point in the initial neighborhood (the neighboring points of the target point). Eigenvalue decomposition is performed on the covariance matrix to obtain eigenvalues ​​and their corresponding eigenvectors. The eigenvector corresponding to the smallest eigenvalue is then used as the normal vector. Take the normal vector The distance difference between the maximum and minimum points in the direction According to the distance difference Calculate the neighborhood angle of the initial neighborhood using the following formula. : .

[0034] Based on neighborhood angle The filter window radius is dynamically adjusted according to the following formula. :

[0035] in, These are empirical coefficients set based on actual needs. They are then adjusted according to the dynamically adjusted filter window radius. Determine the ultimate neighborhood around the target point, and determine the mean of all points within the ultimate neighborhood. and standard deviation Based on the mean and standard deviation Determine the absolute threshold for data evaluation according to the following formula. And the relative evaluation threshold of the data :

[0036]

[0037] Where m is the adjustment coefficient. The feature data corresponding to all points in the ultimate neighborhood are compared with the absolute data evaluation threshold. And the relative evaluation threshold of the data A comparison is made, and noise points are identified and removed based on the comparison results. For example, if the value of the feature data exceeds the absolute evaluation threshold, noise points are removed. The sum is greater than the relative evaluation threshold of the data. If the data is noisy, it is identified as such and removed from the target feature data to obtain the denoised target feature data. The denoised target feature data includes denoised object feature data, denoised electricity consumption feature data, and denoised historical return-home data.

[0038] Furthermore, to improve the prediction accuracy of return-home information, it is first necessary to perform feature cross-enhancement processing on the denoised object feature data, denoised electricity consumption feature data, and denoised historical return-home data. Based on this, step 202 specifically includes: determining the object feature vector corresponding to the object feature data, the electricity consumption feature vector corresponding to the electricity consumption feature data, and the return-home feature vector corresponding to the historical return-home data; performing feature-level cross-enhancement processing on the object feature vector, the electricity consumption feature vector, and the return-home feature vector to obtain a feature cross-enhancement vector; performing element-level cross-enhancement processing on the object feature vector, the electricity consumption feature vector, and the return-home feature vector to obtain an element-level cross-enhancement vector; performing low-order cross-enhancement processing on the object feature vector, the electricity consumption feature vector, and the return-home feature vector to obtain a low-order cross-enhancement vector; and performing enhancement transformation processing on the feature cross-enhancement vector, the element-level cross-enhancement vector, and the low-order cross-enhancement vector to obtain the return-home cross-enhancement vector.

[0039] Specifically, feature extraction models (such as CNN models) are used to extract object feature vectors corresponding to object feature data, electricity consumption feature vectors corresponding to electricity consumption feature data, and return-home feature vectors corresponding to historical return-home data. Then, to fully utilize the relationships between data, extract more latent features, and simultaneously handle both high-order and low-order data to make data utilization more efficient and the resulting predictions more accurate, meeting the needs of practical applications, cross-enhancement processing is required for the object feature vectors, electricity consumption feature vectors, and return-home feature vectors. A specific cross-enhancement processing method is as follows: if the object feature vector is (a1, a2), the electricity consumption feature vector is (b1, b2), and the return-home feature vector is (c1, c2), the specific cross-enhancement processing method includes: performing feature-level cross-enhancement between different feature vectors, that is, after performing a Hadamard product on all elements of the vectors, a convolution transformation is performed under a certain weight w1 to obtain the feature cross-enhancement vector f(w1×(a1×b1×c1, a2×...). Simultaneously, element-level cross-enhancement is performed on all feature vector data. This involves performing a Hadamard product on each element of the vectors, assigning different weight values ​​w2 and w3 to each product, and then performing a linear transformation to obtain the element-level cross-enhancement vector f(w2×a1×b1×c1, w3×a2×b2×c2). Furthermore, low-order cross-enhancement is performed on all feature vectors, and the result of the cross-enhancement is assigned a weight coefficient w4, followed by a linear transformation to obtain the low-order cross-enhancement vector f(w4(a1,a2,b1,b2,c1,c2)). Finally, the feature cross-enhancement vector, element cross-enhancement vector, and low-order cross-enhancement vector are transformed, such as by horizontal concatenation, to obtain the return-to-home cross-enhancement vector. It should be noted that the above examples are merely illustrative and do not limit the scope of this application. Therefore, by performing cross-enhancement processing on the object feature vector, electricity consumption feature vector, and return-home feature vector, different features can be automatically or explicitly combined to generate new feature combinations. These combined features may contain complex nonlinear relationships between the original features, enabling the model to capture more refined and richer information in the data. In other words, it can make full use of the relationships between various data, extract more latent features, and take into account both high-order and low-order processing, making the data utilization more efficient and the subsequent return-home prediction results more accurate, meeting the needs of practical application scenarios.

[0040] 203. Construct a pre-defined initial return-home information prediction model.

[0041] Specifically, the model results of the preset initial return-home information prediction model can have the same model structure as the preset return-home information prediction model.

[0042] 204. Obtain the sample dataset, which includes sample object feature data, sample electricity consumption feature data, and sample historical return data of sample objects with holiday return information tags.

[0043] In this embodiment of the invention, in order to improve the training accuracy of the model, it is first necessary to reasonably obtain a sample dataset. Based on this, step 204 specifically includes: using a preset population inflow and outflow model to identify users who have experienced population outflow and whose homes are currently vacant during a preset period; identifying users who regularly return home during the same historical period based on historical data, wherein the historical data refers to the activity range of users during historical holidays of the same type as the target holiday; extracting the intersection users between the users to be identified and the returning users as the sample objects, and obtaining the sample object feature data, sample electricity consumption feature data, and sample historical return data of the sample objects, and the sample dataset is composed of the sample objects and their corresponding sample object feature data, sample electricity consumption feature data, and sample historical return data.

[0044] Specifically, to improve the recognition accuracy of the preset population inflow and outflow model, it is first necessary to train and construct the preset population inflow and outflow model. Based on this, the method includes: constructing a preset initial population inflow and outflow model and obtaining a sample population dataset, wherein the sample population dataset includes video stream data of sample users with population outflow labels and household vacancy status labels; dividing the sample population dataset into a population training set and a population test set; training the preset initial population inflow and outflow model using the population training set and testing the trained preset initial population inflow and outflow model using the population test set; and finally, using the trained preset initial population inflow and outflow model that meets the test conditions as the preset population inflow and outflow model. Further, acquiring regional video surveillance data of electricity users in a certain area or certain areas, inputting the regional video surveillance data into the preset population inflow and outflow model to identify the outflow of users and the vacancy status of the households where users reside; and, based on the identification results, identifying users in the prediction area (a certain area or certain areas) whose population is outflowing and whose households are currently vacant. Meanwhile, if the target holiday is the National Day holiday, then users who regularly return home during historical National Day holidays in the predicted area are obtained. Then, the users at the intersection of the users to be identified and the returning users are used as sample objects. Sample object feature data, sample electricity consumption feature data, and sample historical return home data of the sample objects are obtained from the power grid database. Finally, the sample dataset is composed of the sample objects and their corresponding sample object feature data, sample electricity consumption feature data, sample historical return home data, and the return home situation of the sample objects.

[0045] 205. Divide the sample dataset into a training set and a test set. Use the training set to train the preset initial return home information prediction model, and use the test set to test the trained preset initial return home information prediction model. Finally, use the trained preset initial return home information prediction model that meets the test conditions as the preset return home information prediction model.

[0046] Specifically, during model training, a pre-defined initial model for predicting homecoming information is first constructed, followed by the acquisition of a sample dataset. The dataset must contain all necessary files. The data is then converted to a format understandable by the pre-defined initial model for predicting homecoming information. Finally, the model is trained and tested. Specifically, the dataset can be divided first: using randomness or a specific strategy (such as stratified sampling), the sample dataset is divided into a training set and a test set. The model is then trained using the training set, and tested using the test set to evaluate its performance on unseen data. Precision, recall, and other metrics on the test set are calculated and recorded. If the model performance does not meet requirements, it can return to the training phase for further iterations or adjustments. This process yields a pre-defined model for predicting homecoming information that meets the requirements.

[0047] 206. Input the cross-enhancement vector of returning home into the preset returning home information prediction model to predict the returning home information and obtain the potential returning home results of the target object during the target holiday.

[0048] In this embodiment of the invention, to further improve the prediction accuracy of homecoming information, step 206 specifically includes: determining multiple homecoming information groups and determining the centroid vector corresponding to each homecoming information group; based on the similarity between the homecoming cross-enhancement vector and each centroid vector, determining the target homecoming information group to which the target object belongs in each homecoming information group; determining the group homecoming information of the target homecoming information group, and inputting the group homecoming information and the homecoming cross-enhancement vector into a preset homecoming information prediction model to predict homecoming information, thereby obtaining the potential homecoming result of the target object during the target holiday.

[0049] Among them, multiple groups of people returning home refer to groups of users with different characteristics, different electricity consumption characteristics, and different historical data on returning home.

[0050] Specifically, for each electricity user within each group of people returning home, feature cross-enhancement processing is performed on the user's object feature data, electricity consumption feature data, and historical return data to obtain a feature cross-enhancement vector. This vector is then fused to obtain the centroid vector of each group. Next, the similarity between the target object's return cross-enhancement vector and the centroid vector of each group is calculated, and the group with the highest similarity is identified as the target group. The past return-home situations of each user within the target group are determined, and based on this, the group's return probability is calculated. This group return probability is used as the group return information. Finally, the group return probability and the return cross-enhancement vector are input into a pre-defined return information prediction model to predict the potential return probability of the target object during the target holiday. This invention predicts the return-home information of a target object by introducing the target return-home information group to which the target object belongs. The group data, by aggregating the behavioral patterns of similar users, can provide richer contextual information and avoid the problem of limited additional data for a single target object, thereby improving the prediction accuracy of the target object's return-home information.

[0051] In another embodiment of the present invention, the method for predicting the return-home information of a target object further includes: acquiring network behavior data and communication data of the target object; performing data fusion processing on the network behavior data and the communication data to obtain a fused feature vector; horizontally concatenating the fused feature vector and the return-home cross-enhancement vector to obtain a concatenated feature vector; and inputting the concatenated feature vector into the preset return-home information prediction model to predict the return-home information, thereby obtaining the potential return-home result of the target object during the target holiday.

[0052] Among them, network behavior data refers to the interactive behavior data of the target object on the Internet platform, including location check-in data, topics related to returning home (such as "ticket grabbing strategy" "countdown to returning home") or emotional tendencies (such as "looking forward to returning home" "anxious about traffic jams") posted by the object; communication data includes: the switching records of the object's mobile phone between different base stations, which can reconstruct its movement path, and the identification of the user's work location through long-term base station location data, the frequency of calls between the user and hometown numbers (such as parents and relatives), and the text messages received by the user from their hometown (such as "picking you up at the station" "New Year's Eve dinner is ready"), etc.

[0053] Specifically, the method for performing data fusion processing on the network behavior data and the communication data to obtain a fused feature vector includes: constructing a data feature matrix based on the data feature vectors corresponding to the network behavior data and the communication data, and constructing a scatter matrix corresponding to the data feature matrix; performing eigenvalue decomposition on the scatter matrix to obtain eigenvalues ​​and eigenvectors corresponding to the eigenvalues, and selecting multiple target feature vectors as principal axis feature vectors of the distribution of network behavior data and communication data based on the magnitude of the eigenvalues, and performing feature fusion on each principal axis feature vector to obtain a principal axis fused feature vector; determining the feature projection direction corresponding to the network behavior data and the communication data, and projecting the data feature vectors according to the feature projection direction to obtain projected data feature vectors; selecting principal component mapping features from the projected data feature vectors based on a preset threshold, and performing feature fusion on the principal component mapping features to obtain a mapping fused feature vector; and performing weighted fusion on the principal axis fused feature vector and the mapping fused feature vector to obtain the fused feature vector.

[0054] Specifically, the scatter matrix is ​​decomposed into eigenvalues. The decomposition method is as follows: First, the eigenvector groups of the scatter matrix are determined, and the corresponding eigenvalues ​​are calculated based on these eigenvector groups. For example, if there are k eigenvector groups, there are k corresponding eigenvalues, each with its own eigenvector. These k eigenvalues ​​are then sorted from largest to smallest to obtain sorted eigenvalues. Next, the top n (preset number) eigenvalues ​​are selected from the sorted eigenvalues, and their corresponding eigenvectors are determined as the principal axis eigenvectors. Finally, feature fusion is performed on each principal axis eigenvector, such as through horizontal concatenation, to obtain the principal axis fused eigenvector.

[0055] Further, the mean vector of the data feature vectors is determined; based on the data feature vectors and their corresponding mean vectors, the intra-class scatter matrix is ​​determined; based on the mean vectors, the inter-class scatter matrix is ​​determined; based on the intra-class scatter matrix and the inter-class scatter matrix, the fusion scatter matrix is ​​determined; the fusion scatter matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalues ​​of the scatter matrix and their corresponding eigenvectors; the eigenvalue of the largest scatter matrix is ​​determined among the eigenvalues ​​of the scatter matrix, and the feature projection direction is determined based on the eigenvector corresponding to the eigenvalue of the largest scatter matrix. The intra-class scatter matrix is ​​determined according to the following formula:

[0056] Where Sw1 represents the within-class scatter matrix, x iLet μ1 represent the element in the data feature vector, c1 represent the mean vector, and c1 represent the data category. The within-class scatter matrix can then be calculated using the formula above. Further, the between-class scatter matrix S is determined according to the following formula. b :

[0057] Where μ2 represents the baseline value of the mean vector, and the fusion divergence matrix is ​​determined based on the intra-class divergence matrix and the inter-class divergence matrix. The fusion divergence matrix is ​​then decomposed into eigenvalues ​​to obtain each eigenvalue and its corresponding eigenvector. The largest eigenvalue is identified, and the direction of the eigenvector corresponding to this largest eigenvalue is determined as the feature projection direction. Further, the data feature vectors are projected along this feature projection direction, that is, the data feature vectors corresponding to network behavior data and communication data are projected onto the line containing the eigenvector corresponding to the largest eigenvalue, resulting in projected data feature vectors. Then, features greater than a preset threshold (set according to actual needs) are selected from the projected data feature vectors as principal component mapping features. Finally, the principal component mapping features are horizontally concatenated to obtain the mapping fusion feature vector. Finally, the principal axis fusion feature vector and the mapping fusion feature vector are weighted and fused to obtain the fusion feature vector.

[0058] Furthermore, the fused feature vector and the cross-enhanced return-home vector are horizontally concatenated. Finally, a preset return-home information prediction model is used to analyze the concatenated features, and the potential return-home outcome, i.e., the potential return-home probability, is predicted based on the analysis results. This embodiment of the invention predicts the return-home situation of the target object by comprehensively analyzing multi-dimensional information such as the target object's network behavior data, communication data, object feature data, electricity consumption feature data, and historical return-home data, which can improve the prediction accuracy of return-home information for electricity users (target objects).

[0059] According to another method for predicting homecoming information provided by the present invention, compared with the current method of judging a user's homecoming situation in future holidays solely based on the user's daily electricity consumption during past holidays, the present invention, in response to the homecoming information prediction signal of the target object during the target holiday, acquires the target object's object feature data, electricity consumption feature data, and historical homecoming data; then, it performs feature cross-enhancement processing on the object feature data, electricity consumption feature data, and historical homecoming data to obtain a homecoming cross-enhancement vector; finally, it inputs the homecoming cross-enhancement vector into a preset homecoming information prediction model to predict homecoming information, thereby obtaining the potential homecoming result of the target object during the target holiday. Thus, by comprehensively analyzing multi-dimensional information such as object feature data, electricity consumption feature data, and historical homecoming data to determine the target object's homecoming situation, the prediction accuracy of the object's homecoming situation can be improved; by performing feature cross-enhancement processing on the object feature data, electricity consumption feature data, and historical homecoming data, more latent features can be extracted, and the relationships between data can be fully utilized, making the subsequent prediction results more accurate; by using a preset homecoming information prediction model to predict the object's homecoming situation, the prediction accuracy and efficiency of homecoming information can be further improved.

[0060] Furthermore, as Figure 1 In specific implementation, embodiments of the present invention provide a device for predicting return-to-hometown information, such as... Figure 3 As shown, the device includes: an acquisition unit 31, a cross-enhancement unit 32, and a homecoming prediction unit 33.

[0061] The acquisition unit 31 can be used to acquire the target object's object feature data, electricity consumption feature data, and historical return home data in response to the target object's return home information prediction signal during the target holiday.

[0062] The cross-enhancement unit 32 can be used to perform feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return-home data to obtain the return-home cross-enhancement vector.

[0063] The homecoming prediction unit 33 can be used to input the homecoming cross-enhancement vector into a preset homecoming information prediction model to predict homecoming information and obtain the potential homecoming result of the target object during the target holiday.

[0064] In specific application scenarios, in order to construct a pre-defined prediction model for return-to-hometown information, such as Figure 4 As shown, the device also includes a construction unit 34.

[0065] The construction unit 34 can be used to construct a preset initial return-home information prediction model; obtain a sample dataset, wherein the sample dataset includes sample object feature data, sample electricity consumption feature data, and sample historical return-home data of sample objects with holiday return-home information tags; divide the sample dataset into a training set and a test set, use the training set to train the preset initial return-home information prediction model, and use the test set to test the trained preset initial return-home information prediction model, and finally use the trained preset initial return-home information prediction model that meets the test conditions as the preset return-home information prediction model.

[0066] In specific application scenarios, in order to obtain sample datasets, the construction unit 34 includes an identification module 341 and an extraction module 342.

[0067] The identification module 341 can be used to identify users who have experienced population outflow and whose homes are currently vacant during a preset period using a preset population inflow and outflow model.

[0068] The identification module 341 can also be used to identify users who regularly return home during the same historical period based on historical data, wherein the historical data refers to the activity range of users during historical holidays of the same type as the target holiday.

[0069] The extraction module 342 can be used to extract the intersection users between the user to be identified and the user returning home as the sample object, and to obtain the sample object feature data, sample electricity consumption feature data, and sample historical return home data of the sample object. The sample dataset is composed of the sample object and its corresponding sample object feature data, sample electricity consumption feature data, and sample historical return home data.

[0070] In specific application scenarios, in order to perform feature cross-enhancement processing on the data, the cross-enhancement unit 32 includes a first determination module 321, a cross-enhancement module 322, and an enhancement transformation module 323.

[0071] The first determining module 321 can be used to determine the object feature vector corresponding to the object feature data, the electricity consumption feature vector corresponding to the electricity consumption feature data, and the return home feature vector corresponding to the historical return home data, respectively.

[0072] The cross-enhancement module 322 can be used to perform feature-level cross-enhancement processing on the object feature vector, the electricity consumption feature vector, and the return-home feature vector to obtain a feature cross-enhancement vector.

[0073] The cross-enhancement module 322 can also be used to perform element-level cross-enhancement processing on the object feature vector, the electricity consumption feature vector, and the return-home feature vector to obtain an element-level cross-enhancement vector.

[0074] The cross-enhancement module 322 can also be used to perform low-order cross-enhancement processing on the object feature vector, the electricity consumption feature vector, and the return-home feature vector to obtain a low-order cross-enhancement vector.

[0075] The enhancement transformation module 323 can be used to perform enhancement transformation processing on the feature cross enhancement vector, the element cross enhancement vector, and the low-order cross enhancement vector to obtain the homecoming cross enhancement vector.

[0076] In specific application scenarios, in order to process noise in object feature data, electricity consumption feature data, and historical return home data, the device also includes a noise reduction unit 35.

[0077] The denoising unit 35 can be used to take any one of the object feature data, the electricity consumption feature data, and the historical return-home data as a target feature data, and take any one of the target feature data as a target point, determine the initial neighborhood corresponding to the target point, and form a covariance matrix from the neighborhood points in the initial neighborhood; perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors corresponding to the eigenvalues, take the eigenvector corresponding to the minimum eigenvalue as the normal vector, determine the distance difference between the maximum and minimum points in the direction of the normal vector, and determine the target point based on the distance difference. The initial neighborhood angle is determined; based on the neighborhood angle, the initial neighborhood is adjusted to obtain the ultimate neighborhood corresponding to the target point; the mean and standard deviation of all points within the ultimate neighborhood are determined; based on the mean and standard deviation, an absolute data evaluation threshold and a relative data evaluation threshold are determined respectively; based on the absolute data evaluation threshold and the relative data evaluation threshold, noise judgment is performed on the target feature data represented by the target point; based on the noise judgment result, noise points are identified in the target feature data, and the noise points in the target feature data are removed to obtain the denoised target feature data.

[0078] In specific application scenarios, in order to determine the cross-enhancement vector for returning home, the cross-enhancement unit 32 can be used to perform feature cross-enhancement processing on the denoised object feature data, the denoised electricity consumption feature data, and the denoised historical returning home data to obtain the cross-enhancement vector for returning home.

[0079] In specific application scenarios, in order to predict the return home information of the target, the return home prediction unit 33 includes a second determination module 331 and a prediction module 332.

[0080] The second determining module 331 can be used to determine multiple groups of people returning home and to determine the centroid vector corresponding to each group of people returning home.

[0081] The second determining module 331 can also be used to determine the target homecoming information group to which the target object belongs in each homecoming information group based on the similarity between the homecoming cross-enhancement vector and each centroid vector.

[0082] The prediction module 332 can be used to determine the group return information of the target return information group, and input the group return information and the return cross-enhancement vector into the preset return information prediction model to predict the return information, so as to obtain the potential return results of the target object during the target holiday.

[0083] In specific application scenarios, in order to predict return-home information, the return-home prediction unit 33 also includes an acquisition module 333 and a fusion module 334.

[0084] The acquisition module 333 can be used to acquire the network behavior data and communication data of the target object.

[0085] The fusion module 334 can be used to perform data fusion processing on the network behavior data and the communication data to obtain a fused feature vector.

[0086] The prediction module 332 can also be used to horizontally concatenate the fused feature vector and the homecoming cross-enhancement vector to obtain a concatenated feature vector, and input the concatenated feature vector into the preset homecoming information prediction model to predict homecoming information, thereby obtaining the potential homecoming result of the target object during the target holiday.

[0087] It should be noted that other corresponding descriptions of the functional modules involved in the homecoming information prediction device provided in this embodiment of the invention can be found in [reference]. Figure 1 The corresponding description of the method shown will not be repeated here.

[0088] Based on the above, Figure 1 Accordingly, this embodiment of the invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the following steps: in response to a prediction signal of a target object's return home information during a target holiday, acquiring object feature data, electricity consumption feature data, and historical return home data of the target object; performing feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return home data to obtain a return home cross-enhancement vector; inputting the return home cross-enhancement vector into a preset return home information prediction model to predict return home information, thereby obtaining the potential return home result of the target object during the target holiday.

[0089] Based on the above, Figure 1 The method shown and as Figure 3 The embodiment of the device shown in the invention also provides a physical structure diagram of a computer device, such as... Figure 5 As shown, the computer device includes: a processor 41, a memory 42, and a computer program stored in the memory 42 and executable on the processor. Both the memory 42 and the processor 41 are mounted on a bus 43. When the processor 41 executes the program, it performs the following steps: in response to a prediction signal of a target object's return-home information during a target holiday, it acquires the target object's object feature data, electricity consumption feature data, and historical return-home data; it performs feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return-home data to obtain a return-home cross-enhancement vector; and it inputs the return-home cross-enhancement vector into a preset return-home information prediction model to predict the return-home information, thereby obtaining the potential return-home result of the target object during the target holiday.

[0090] Through the technical solution of this invention, in response to the return-home information prediction signal of a target object during a target holiday, the invention acquires the target object's object feature data, electricity consumption feature data, and historical return-home data; then, it performs feature cross-enhancement processing on the object feature data, electricity consumption feature data, and historical return-home data to obtain a return-home cross-enhancement vector; finally, it inputs the return-home cross-enhancement vector into a preset return-home information prediction model to predict the return-home information, thereby obtaining the potential return-home result of the target object during the target holiday. Thus, by comprehensively analyzing multi-dimensional information such as object feature data, electricity consumption feature data, and historical return-home data to determine the target object's return-home situation, the prediction accuracy of the object's return-home situation can be improved; by performing feature cross-enhancement processing on the object feature data, electricity consumption feature data, and historical return-home data, more latent features can be extracted, and the relationships between data can be fully utilized, making the subsequent prediction results more accurate; by using a preset return-home information prediction model to predict the object's return-home situation, the prediction accuracy and efficiency of return-home information can be further improved.

[0091] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0092] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting information about people returning to their hometowns, characterized in that, include: In response to the prediction signal of the target object's return home information during the target holiday, the target object's object characteristic data, electricity consumption characteristic data, and historical return home data are acquired; The object feature data, the electricity consumption feature data, and the historical return-home data are subjected to feature cross-enhancement processing to obtain the return-home cross-enhancement vector; The cross-enhancement vector for returning home is input into a preset return home information prediction model to predict return home information, thereby obtaining the potential return home results of the target object during the target holiday.

2. The method according to claim 1, characterized in that, Before inputting the return-home cross-enhancement vector into a preset return-home information prediction model to predict return-home information and obtain the return-home result of the target object, the method further includes: Construct a pre-defined initial return-home information prediction model; Obtain a sample dataset, wherein the sample dataset includes sample object feature data, sample electricity consumption feature data, and sample historical return home data of sample objects with holiday return home information tags; The sample dataset is divided into a training set and a test set. The training set is used to train the preset initial return home information prediction model, and the test set is used to test the trained preset initial return home information prediction model. Finally, the trained preset initial return home information prediction model that meets the test conditions is taken as the preset return home information prediction model.

3. The method according to claim 2, characterized in that, The acquisition of the sample dataset includes: Use a preset population inflow and outflow model to identify users who have experienced population outflow during a preset period and whose homes are currently vacant; Based on historical data from the same period, identify users who return home in a regular pattern during the same period in history, wherein the historical data refers to the activity range of users during historical holidays of the same type as the target holiday; The users that intersect with the users to be identified and the users returning home are extracted as the sample objects. The sample object feature data, sample electricity consumption feature data, and sample historical return home data of the sample objects are obtained. The sample dataset is composed of the sample objects and their corresponding sample object feature data, sample electricity consumption feature data, and sample historical return home data.

4. The method according to claim 1, characterized in that, The step of performing feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return-home data to obtain the return-home cross-enhancement vector includes: Determine the object feature vector corresponding to the object feature data, the electricity consumption feature vector corresponding to the electricity consumption feature data, and the return home feature vector corresponding to the historical return home data, respectively. The object feature vector, the electricity consumption feature vector, and the return-home feature vector are subjected to feature-level cross-enhancement processing to obtain the feature cross-enhancement vector; Element-level cross-enhancement processing is performed on the object feature vector, the electricity consumption feature vector, and the return-home feature vector to obtain the element-level cross-enhancement vector; The object feature vector, the electricity consumption feature vector, and the return-home feature vector are subjected to low-order cross-enhancement processing to obtain a low-order cross-enhancement vector. The feature cross-enhancement vector, the element cross-enhancement vector, and the low-order cross-enhancement vector are subjected to enhancement transformation to obtain the homecoming cross-enhancement vector.

5. The method according to claim 1, characterized in that, Before performing feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return-home data to obtain the return-home cross-enhancement vector, the method further includes: Each of the object feature data, the electricity consumption feature data, and the historical return-home data can be used as a target feature data. Take any data point in the target feature data as a target point, determine the initial neighborhood corresponding to the target point, and form a covariance matrix from the neighborhood points in the initial neighborhood. The covariance matrix is ​​decomposed into eigenvalues ​​and corresponding eigenvectors. The eigenvector corresponding to the smallest eigenvalue is used as the normal vector. The distance difference between the maximum and minimum points in the direction of the normal vector is determined. Based on the distance difference, the neighborhood angle of the initial neighborhood is determined. Based on the neighborhood angle, the initial neighborhood is adjusted to obtain the ultimate neighborhood corresponding to the target point. The mean and standard deviation of all points within the ultimate neighborhood are determined. Based on the mean and standard deviation, the absolute data evaluation threshold and the relative data evaluation threshold are determined respectively. Based on the absolute evaluation threshold and the relative evaluation threshold of the data, noise judgment is performed on the target feature data represented by the target point. Based on the noise judgment result, noise points are identified in the target feature data and the noise points in the target feature data are removed to obtain the denoised target feature data. The step of performing feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return-home data to obtain the return-home cross-enhancement vector includes: The denoised object feature data, the denoised electricity consumption feature data, and the denoised historical return-home data are subjected to feature cross-enhancement processing to obtain the return-home cross-enhancement vector.

6. The method according to claim 1, characterized in that, The step of inputting the return-home cross-enhancement vector into a preset return-home information prediction model to predict return-home information and obtain the potential return-home results of the target object during the target holiday includes: Multiple groups of people returning to their hometowns are identified, and the centroid vector corresponding to each group is determined. Based on the similarity between the cross-enhancing vector for returning home and each centroid vector, the target returning home information group to which the target object belongs is determined in each returning home information group; Determine the group return information of the target group, and input the group return information and the return cross-enhancement vector into a preset return information prediction model to predict the return information, thereby obtaining the potential return results of the target object during the target holiday.

7. The method according to claim 1, characterized in that, The step of inputting the return-home cross-enhancement vector into a preset return-home information prediction model to predict return-home information and obtain the potential return-home results of the target object during the target holiday includes: Obtain the network behavior data and communication data of the target object; The network behavior data and the communication data are fused to obtain a fused feature vector; The fused feature vector and the homecoming cross-enhancement vector are horizontally concatenated to obtain a concatenated feature vector. The concatenated feature vector is then input into the preset homecoming information prediction model to predict homecoming information, thereby obtaining the potential homecoming result of the target object during the target holiday.

8. A device for predicting information on people returning home, characterized in that, include: The acquisition unit is used to acquire the target object's object feature data, electricity consumption feature data, and historical return home data in response to the target object's return home information prediction signal during the target holiday; The cross-enhancement unit is used to perform feature cross-enhancement processing on the object feature data, the electricity consumption feature data, and the historical return-home data to obtain the return-home cross-enhancement vector. The homecoming prediction unit is used to input the homecoming cross-enhancement vector into a preset homecoming information prediction model to predict homecoming information and obtain the potential homecoming results of the target object during the target holiday.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images