Energy consumption data prediction method, device, medium and terminal based on feature screening

Abstract

The present application discloses a method and device, medium, and terminal for predicting energy consumption data based on feature screening, which relates to the field of spatiotemporal characteristic analysis of energy consumption, and its main purpose is to solve the problem of low prediction accuracy of future energy consumption behavior data of users. It includes: obtaining energy consumption behavior data and environmental parameter sets of all users in the target area; determining the typical environmental parameter set of each user according to the maximum correlation minimum redundancy index; determining the user's correlation coefficient vector according to the typical environmental parameter set; clustering users into multiple user groups according to the correlation coefficient vector; integrating the typical environmental parameter sets of each user in the user group to obtain the environmental parameter indicator vector of the user group; calculating the feature screening traversal solution index of the environmental parameter indicator vector under different values, and determining the optimal environmental parameter set of the user group according to the maximum value; using the energy consumption behavior data prediction model of each user group, predicting the energy consumption behavior data according to the optimal environmental parameter set and energy consumption behavior data.

Description

Technical Field

[0001] The present application relates to the field of spatiotemporal characteristic analysis of energy consumption, and in particular to a method and device, medium, and terminal for predicting energy consumption data based on feature screening. Background Art

[0002] With the rapid development of new power systems, it has become possible for the load side to participate in system regulation. However, when a large number of flexible loads such as electric vehicles, energy storage, and demand response are connected to the distribution network, due to their scattered and disorderly access characteristics, it becomes difficult to analyze the energy consumption characteristics of users, resulting in a decrease in the accuracy of the prediction of users' future energy consumption behavior data. Accurate prediction results of users' future energy consumption behavior data are of great significance to optimizing the operation of the power system, improving energy utilization efficiency, and ensuring the stable operation of the power system.

[0003] At present, the commonly used analysis methods of user energy consumption characteristics mainly focus on regression method and Kalman filtering method based on statistical theory. However, since regression method and Kalman filtering method only focus on the data itself and ignore the impact of nonlinear problems on user energy consumption characteristics, the analysis of user energy consumption characteristics is unreliable, which leads to a decrease in the accuracy of predicting user future energy consumption behavior data. Summary of the invention

[0004] In view of this, the present application provides a method and device, medium, and terminal for predicting energy consumption data based on feature screening. The main purpose is to solve the problem that the existing analysis of user energy consumption characteristics is unreliable, which leads to reduced accuracy in predicting user future energy consumption behavior data.

[0005] According to one aspect of the present application, a method for predicting energy consumption data based on feature screening is provided, comprising:

[0006] Obtaining energy usage behavior data of all users in the target area within a preset time period and an environmental parameter set of the target area;

[0007] Performing a first random division on the multiple environmental parameters included in the environmental parameter set to generate multiple environmental parameter subsets, wherein the first random division includes the number of environmental parameter subsets, the number of environmental parameters in the environmental parameter subsets, and the types of environmental parameters in the environmental parameter subsets;

[0008] Each of the users is taken as a target user one by one, and the maximum correlation minimum redundancy index between each of the environmental parameter subsets and the energy consumption behavior data of the target user is calculated respectively, and the environmental parameter subset corresponding to the maximum value is screened out from the multiple maximum correlation minimum redundancy indexes corresponding to the target user, and the environmental parameter subset is taken as the typical environmental parameter set of the target user, and the maximum correlation minimum redundancy index is used to characterize the difference between the maximum correlation index and the minimum redundancy index between the environmental parameter subset and the energy consumption behavior data of the user;

[0009] Calculating the correlation coefficient between each environmental parameter included in the typical environmental parameter set and the energy usage behavior data of the target user respectively, combining each of the correlation coefficients to obtain a correlation coefficient vector of the target user, and combining each of the correlation coefficient vectors of the target user to obtain a correlation coefficient matrix between the user and the typical environmental parameter set;

[0010] Performing clustering processing on all users based on the correlation coefficient matrix to obtain multiple user groups, and taking each of the user groups as a target user group;

[0011] Integrate the typical environment parameter sets of each user included in the target user group to obtain an environment parameter indication vector of the target user group, perform a second random division on multiple environment parameters included in the environment parameter indication vector to generate multiple environment parameter indication sub-vectors, wherein the second random division includes the number of environment parameter indication sub-vectors, the number of environment parameters in the environment parameter indication sub-vectors, and the types of environment parameters in the environment parameter indication sub-vectors;

[0012] Calculate the characteristic screening traversal solution index of each of the environmental parameter indication sub-vectors respectively, select the maximum characteristic screening traversal solution index from each of the characteristic screening traversal solution indexes, determine the environmental parameter indication sub-vector corresponding to the maximum characteristic screening traversal solution index as the optimal environmental parameter set of the target user group, and obtain the optimal environmental parameter set of each user group, wherein the characteristic screening traversal solution index is used to measure the degree of adaptability between the environmental parameter indication sub-vector and the energy consumption behavior data of the user group;

[0013] The energy usage behavior data prediction models matching each of the user groups are called respectively, and prediction processing is performed according to the optimal environmental parameter set and energy usage behavior data of each of the user groups to obtain the predicted energy usage behavior data of each of the user groups. The energy usage behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model, and the encoder and the decoder are constructed based on long-short term neural networks.

[0014] Preferably, the calculating of the maximum correlation minimum redundancy index between the environmental parameter subset and the energy usage behavior data of the target user includes:

[0015] Obtaining a correlation coefficient between each environmental parameter included in the environmental parameter subset and the power consumption behavior data of the target user;

[0016] Based on the following formula, the maximum correlation index between the environmental parameter subset and the power consumption behavior data of the target user is calculated according to each of the correlation coefficients:

[0017] ,

[0018] in, Represents a subset of environmental parameters Data on electricity usage behavior of users The maximum correlation index between Represents a subset of environmental parameters The number of environment parameters included, Indicates environmental parameters Energy usage behavior data of users The correlation coefficient between

[0019] Taking any two environmental parameters in the environmental parameter subset as target environmental parameter pairs one by one;

[0020] Dispersing the data points included in the two environmental parameters in the target environmental parameter pair in pairs in a two-dimensional space to obtain a data point set of the target environmental parameter pair;

[0021] According to different division scales, the two-dimensional space is gridded to obtain multiple two-dimensional space grids, and the mutual information index of the data point set in each of the two-dimensional space grids is calculated based on the following formula:

[0022] ,

[0023] in, Represents a collection of data points In two-dimensional space The mutual information indicator in represents the environmental parameters, A represents the interval in the x-axis direction of the two-dimensional space, and B represents the interval in the y-axis direction of the two-dimensional space. Indicates that the data points in the data point set are located in the region The probability of Indicates that the data points in the data point set are located in the region The probability of Indicates that the data points in the data point set are located in the region The probability of express from Take the value in express from Take the value in

[0024] The maximum mutual information index is selected from the plurality of mutual information indexes, and the maximum mutual information index is normalized based on the following formula to obtain the normalized maximum mutual information index.

[0025] ,

[0026] in, Represents the maximum mutual information index of the data point set after normalization in the two-dimensional space grid, Represents the maximum mutual information index of a data point set in a two-dimensional space grid, Represents the number of grids in the two-dimensional space grid in the x-axis direction. Represents the number of grids in the y-axis direction of the two-dimensional space grid;

[0027] Based on the following formula, the maximum information coefficient of the target environment parameter pair is calculated according to the maximum mutual information index after the normalization process:

[0028] ,

[0029] in, Indicates environmental parameters and environmental parameters The maximum information coefficient between represents the mesh complexity parameter;

[0030] Based on the following formula, the minimum redundancy index of the environmental parameter subset is calculated according to the maximum information coefficient of each target environmental parameter pair:

[0031] ,

[0032] in, represents the minimum redundancy index of a subset of environmental parameters;

[0033] The difference between the maximum correlation index and the minimum redundancy index is determined as the maximum correlation minimum redundancy index between the environmental parameter subset and the energy usage behavior data of the target user.

[0034] Preferably, before obtaining the correlation coefficient between each environmental parameter included in the environmental parameter subset and the power usage behavior data of the target user, the method further includes:

[0035] Taking each environmental parameter included in the environmental parameter set as a target environmental parameter one by one, and taking each user included in the target area as a target user one by one;

[0036] Based on the following formula, the correlation coefficient between the target environmental parameters and the target user's energy consumption behavior data is calculated.

[0037] ,

[0038] in, Indicates environmental parameters Energy usage behavior data of users The covariance between Indicates environmental parameters The standard deviation of Represents the user's energy usage behavior data The standard deviation of .

[0039] Preferably, before respectively calling the energy consumption behavior data prediction models matching the respective user groups, the method further comprises:

[0040] Constructing a corresponding initial energy consumption behavior data prediction model for each of the user groups, and using each of the initial energy consumption behavior data prediction models as a target initial energy consumption behavior data prediction model, wherein the initial energy consumption behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model, and the encoder and the decoder are constructed based on a long-short term neural network;

[0041] Obtaining energy usage behavior data of a user group corresponding to the target initial energy usage behavior data prediction model in a first historical period and a historical optimal environmental parameter set;

[0042] Inputting the energy usage behavior data of the first historical period and the historical optimal environmental parameter set into the target initial energy usage behavior data prediction model to obtain predicted energy usage behavior data of the second historical period;

[0043] Performing model training on the target initial energy usage behavior data prediction model based on a loss function between the predicted energy usage behavior data and the actual energy usage behavior data of the second historical period to obtain a target energy usage behavior data prediction model for which model training has been completed;

[0044] The energy usage behavior data prediction models corresponding to the respective user groups that have completed model training are obtained, so as to predict the predicted energy usage behavior data of the respective user groups based on the energy usage behavior data prediction models that have completed model training.

[0045] Preferably, clustering all users based on the correlation coefficient matrix to obtain multiple user groups includes:

[0046] Based on the number of pre-configured user groups, randomly select the number of users as initial cluster centroids;

[0047] Calculating the Euclidean distance between the correlation coefficient vector of each user and the correlation coefficient vector of each user corresponding to the initial clustering centroid, selecting the initial clustering centroid corresponding to the minimum Euclidean distance for each user, and assigning each user to the cluster corresponding to the initial clustering centroid, wherein the correlation coefficient vector is obtained from the correlation coefficient matrix;

[0048] Calculating the center vector of each cluster, and determining the center vector as the updated cluster centroid;

[0049] The Euclidean distance between the correlation coefficient vector of each user and each updated cluster centroid is calculated respectively, and each user is reallocated to the cluster after the cluster centroid is updated, until a preset number of iterations is reached or the center vector of each cluster remains unchanged, thereby obtaining multiple user groups.

[0050] Preferably, before randomly selecting the number of users as initial cluster centroids based on the number of preconfigured user groups, the method further includes:

[0051] Pre-configure multiple user groups;

[0052] Calculate the sum of squared errors within each cluster under different numbers of user groups, and generate a curve showing the sum of squared errors within the cluster changing with the number of user groups;

[0053] Based on the elbow method, the optimal number of user groups is determined according to the curve, so as to select initial cluster centroids based on the optimal number of user groups.

[0054] Preferably, the environmental parameter set includes at least economic development level parameters, resident income parameters, electricity price parameters, temperature parameters, precipitation parameters, wind speed parameters, population density parameters, and date type parameters.

[0055] According to another aspect of the present application, there is provided an energy consumption data prediction device based on feature screening, comprising:

[0056] A data acquisition module, used to acquire energy usage behavior data of all users included in the target area within a preset time period and an environmental parameter set of the target area;

[0057] An environmental parameter subset division module, used for performing a first random division on a plurality of environmental parameters included in the environmental parameter set to generate a plurality of environmental parameter subsets, wherein the first random division includes the number of environmental parameter subsets, the number of environmental parameters in the environmental parameter subsets, and the types of environmental parameters in the environmental parameter subsets;

[0058] A typical environmental parameter set determination module, used to take each of the users as a target user one by one, and respectively calculate the maximum correlation minimum redundancy index between each of the environmental parameter subsets and the energy usage behavior data of the target user, and select the environmental parameter subset corresponding to the maximum value from the multiple maximum correlation minimum redundancy indexes corresponding to the target user, and take the environmental parameter subset as the typical environmental parameter set of the target user, wherein the maximum correlation minimum redundancy index is used to characterize the difference between the maximum correlation index and the minimum redundancy index between the environmental parameter subset and the energy usage behavior data of the user;

[0059] A correlation coefficient matrix determination module, used to respectively calculate the correlation coefficient between each environmental parameter included in the typical environmental parameter set and the energy usage behavior data of the target user, combine each of the correlation coefficients to obtain a correlation coefficient vector of the target user, and combine each of the correlation coefficient vectors of the target user to obtain a correlation coefficient matrix between the user and the typical environmental parameter set;

[0060] A user clustering module, used to perform clustering processing on all users based on the correlation coefficient matrix to obtain multiple user groups, and use each of the user groups as a target user group;

[0061] an environment parameter indication sub-vector division module, used to integrate the typical environment parameter sets of each user included in the target user group to obtain the environment parameter indication vector of the target user group, and perform a second random division on multiple environment parameters included in the environment parameter indication vector to generate multiple environment parameter indication sub-vectors, wherein the second random division includes the number of environment parameter indication sub-vectors, the number of environment parameters in the environment parameter indication sub-vectors, and the types of environment parameters in the environment parameter indication sub-vectors;

[0062] An optimal environmental parameter set determination module is used to respectively calculate the characteristic screening traversal solution index of each of the environmental parameter indication sub-vectors, screen out the maximum characteristic screening traversal solution index from each of the characteristic screening traversal solution indexes, determine the environmental parameter indication sub-vector corresponding to the maximum characteristic screening traversal solution index as the optimal environmental parameter set of the target user group, and obtain the optimal environmental parameter set of each user group, wherein the characteristic screening traversal solution index is used to measure the degree of adaptability between the environmental parameter indication sub-vector and the energy consumption behavior data of the user group;

[0063] The energy usage behavior data prediction module is used to call the energy usage behavior data prediction model matching each of the user groups respectively, perform prediction processing according to the optimal environmental parameter set and energy usage behavior data of each of the user groups, and obtain the predicted energy usage behavior data of each of the user groups. The energy usage behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model, and the encoder and the decoder are constructed based on long-short-term neural networks.

[0064] According to another aspect of the present application, a storage medium is provided, in which at least one executable instruction is stored, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned energy consumption data prediction method based on feature screening.

[0065] According to another aspect of the present application, a terminal is provided, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus;

[0066] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned energy consumption data prediction method based on feature screening.

[0067] By means of the above technical solution, the technical solution provided by the embodiment of the present application has at least the following advantages:

[0068] The present application provides a method and device, medium, and terminal for predicting energy consumption data based on feature screening. Compared with the prior art, the embodiment of the present application screens out a typical environmental parameter set corresponding to each user from the environmental parameter set based on the maximum correlation minimum redundancy index, and then obtains the correlation coefficient vector of each user based on the correlation coefficient combination between each environmental parameter contained in the typical environmental parameter set and the energy consumption behavior data of the corresponding user, divides all users into multiple user groups based on the correlation coefficient vector, and then integrates the typical environmental parameter sets of each user contained in the user group to obtain the environmental parameter indication vector of the user group, and uses the traversal solution method to determine the optimal environmental parameter set of the user group from the environmental parameter indication vector, which fully considers the correlation between the environmental parameters and the user's energy consumption behavior data and the environmental parameters. The redundancy between them is not limited to the attention on the user energy consumption behavior data itself, and the influence of nonlinear problems such as environmental parameters on the user energy consumption behavior data is considered, so that the analysis results of the user energy consumption characteristics are more reliable; further, the energy consumption behavior data prediction model constructed for each user group is composed of an encoder, a decoder, and a global attention sub-model, among which the encoder and the decoder are constructed based on the long-short-term neural network, which utilizes the characteristics of the long-short-term neural network with a longer short-term memory and fully represents the changes in the time series, and uses the global attention sub-model to consider all hidden layers and learn the weights of different environmental parameters, thereby improving the prediction model's attention to important environmental parameters and reducing the probability of gradual accumulation of errors with iterations, thereby improving the accuracy of the energy consumption behavior data prediction results.

[0069] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0071] Figure 1 A flow chart of an energy consumption data prediction method based on feature screening provided in an embodiment of the present application is shown;

[0072] Figure 2 A calculation flow chart of the maximum relevant minimum redundancy index provided in an embodiment of the present application is shown;

[0073] Figure 3A flowchart of constructing an energy consumption behavior data prediction model provided by an embodiment of the present application is shown;

[0074] Figure 4 A schematic diagram of the S2S framework structure provided in an embodiment of the present application is shown;

[0075] Figure 5 A data processing schematic diagram of the global attention sub-model provided in an embodiment of the present application is shown;

[0076] Figure 6 A flowchart of user clustering provided by an embodiment of the present application is shown;

[0077] Figure 7 A block diagram of a device for predicting energy consumption data based on feature screening provided in an embodiment of the present application is shown;

[0078] Figure 8 A schematic diagram of the structure of a terminal provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0079] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0080] The present application embodiment provides a method for predicting energy consumption data based on feature screening, such as Figure 1 As shown, the method includes:

[0081] 101. Obtain energy usage behavior data of all users in a target area within a preset time period and an environmental parameter set of the target area.

[0082] Among them, the environmental parameter set at least includes economic development level parameters, resident income parameters, electricity price parameters, temperature parameters, precipitation parameters, wind speed parameters, population density parameters, date type parameters, etc., and other environmental parameters can be added according to actual conditions. The above environmental parameters are all environmental parameter sequences; the target area can be the area under the jurisdiction of a power supply station, or a certain administrative district or a certain city, etc.; the preset time can be the previous 24 hours, a week, a month, etc. at the current statistical moment; the energy consumption behavior data can be power load data, etc. In the embodiment of the present application, the current execution end can be an energy consumption behavior data prediction unit of the power distribution system, which is used to predict the power load demand of the target area within a preset time in the future, so as to reserve the power distribution quota according to the demand.

[0083] It should be noted that the energy consumption behavior data of each user can be normalized based on the maximum and minimum value formula in advance, and the energy consumption behavior data can be converted into data in the range of [0,1] to reduce the amount of data calculation. The maximum and minimum value formula is expressed as the following formula

[0084] ,

[0085] in, represents the normalized energy consumption behavior data, Represents the user's energy usage behavior data, Indicates the user's minimum energy consumption behavior data, Indicates the maximum energy consumption behavior data of the user. At the same time, the character type environment parameters in the environment parameter set can be converted into numerical type environment parameters in advance. For example, the date type parameters can be converted into numerical values ​​from 1 to 365 in sequence.

[0086] 102. Perform a first random division on a plurality of environmental parameters included in the environmental parameter set to generate a plurality of environmental parameter subsets.

[0087] Wherein, the first random division includes the number of environmental parameter subsets, the number of environmental parameters in the environmental parameter subsets, and the types of environmental parameters in the environmental parameter subsets. Exemplarily, the environmental parameter set includes the following environmental parameters: economic development level parameter, resident income parameter, electricity price parameter, temperature parameter, precipitation parameter, wind speed parameter, population density parameter, date type parameter, the number of environmental parameter subsets obtained by the first random division is random, the number of environmental parameters in the environmental parameter subsets is random, and the types of environmental parameters in the environmental parameter subsets are also random, such as environmental parameter subset 1 [economic development level parameter, resident income parameter], environmental parameter subset 2 [economic development level parameter, electricity price parameter]... environmental parameter subset 11 [resident income parameter, electricity price parameter], environmental parameter subset 12 [resident income parameter, temperature parameter]... environmental parameter subset 21 [economic development level parameter, resident income parameter, electricity price parameter]... environmental parameter subset 31 [economic development level parameter, resident income parameter, electricity price parameter, temperature parameter], etc.

[0088] 103. Take each user as the target user one by one, and calculate the maximum correlation minimum redundancy index between each environmental parameter subset and the energy consumption behavior data of the target user respectively, and select the environmental parameter subset corresponding to the maximum value from the multiple maximum correlation minimum redundancy indicators corresponding to the target user, and use the environmental parameter subset as the typical environmental parameter set of the target user.

[0089] Among them, the maximum correlation minimum redundancy index is used to characterize the difference between the maximum correlation index and the minimum redundancy index between the environmental parameter subset and the user's energy consumption behavior data; the maximum correlation index between the environmental parameter subset and the user's energy consumption behavior data is used to characterize the mean of the sum of the correlation coefficients between each environmental parameter included in the environmental parameter subset and the user's energy consumption behavior data; the minimum redundancy index is used to characterize the degree of redundancy between each environmental parameter included in the environmental parameter subset. In an embodiment of the present application, a typical environmental parameter set corresponding to each user is screened out from the environmental parameter set based on the maximum correlation minimum redundancy index, fully considering the correlation between the environmental parameters and the user's energy consumption behavior data and the redundancy between the environmental parameters, so that the typical environmental parameter set can more accurately describe the energy consumption characteristics of the user.

[0090] 104. Calculate the correlation coefficients between each environmental parameter included in the typical environmental parameter set and the energy usage behavior data of the target user, combine the correlation coefficients to obtain the correlation coefficient vector of the target user, and combine the correlation coefficient vectors of each target user to obtain the correlation coefficient matrix between the user and the typical environmental parameter set.

[0091] In the embodiment of the present application, after determining the typical environmental parameter set of the target user, the correlation coefficient between each environmental parameter included in the typical environmental parameter set and the energy consumption behavior data of the target user can be calculated based on the correlation coefficient calculation formula. The correlation coefficient calculation formula is expressed as the following formula

[0092] ,

[0093] in, Indicates environmental parameters Energy usage behavior data of users The correlation coefficient between Indicates environmental parameters Energy usage behavior data of users The covariance between Indicates environmental parameters The standard deviation of Represents the user's energy usage behavior data The standard deviation of .

[0094] Furthermore, the correlation coefficients are combined to obtain the correlation coefficient vector of the target user:

[0095] ,

[0096] in, represents the correlation coefficient vector of user n, Indicates environmental parameters Energy usage behavior data of user n The correlation coefficient between .

[0097] Furthermore, the correlation coefficient vectors of each target user are combined to obtain the correlation coefficient matrix between the user and the typical environment parameter set:

[0098] ,

[0099] in, H Represents the correlation coefficient matrix.

[0100] 105. Perform clustering processing on all users based on the correlation coefficient matrix to obtain multiple user groups, and use each user group as a target user group.

[0101] It is understandable that, since the correlation coefficient vector of each user is recorded in the correlation coefficient matrix, the correlation coefficient vector records the correlation coefficient between each environmental parameter contained in the typical environmental parameter set corresponding to the user and the energy consumption behavior data of the user, and the typical environmental parameter set is used to describe the energy consumption characteristics of the user, therefore, clustering based on the correlation coefficient vector of each user as the data basis can effectively improve the accuracy of the clustering results, so that the similarity between the users included in each user group is higher, thereby improving the accuracy of the prediction results. In the embodiment of the present application, the correlation coefficient matrix is ​​used as the input data for clustering processing, and multiple user groups after clustering are output. Further, each user group is taken as the target user group, that is, the target object of the subsequent steps, and the subsequent steps are performed one by one.

[0102] 106. Integrate the typical environment parameter sets of each user included in the target user group to obtain an environment parameter indication vector of the target user group, and perform a second random division on multiple environment parameters included in the environment parameter indication vector to generate multiple environment parameter indication sub-vectors.

[0103] The second random division includes the number of environment parameter indication sub-vectors, the number of environment parameters in the environment parameter indication sub-vectors, and the types of environment parameters in the environment parameter indication sub-vectors.

[0104] In the embodiment of the present application, the environmental parameter indication vector can be expressed as the following formula:

[0105] ,

[0106] in, represents the environmental parameter indicator vector, ,when When , it indicates the environmental parameters Existence, when When , it indicates the environmental parameters Does not exist.

[0107] Furthermore, the environmental parameter indicator vector The multiple environmental parameters included are subjected to a second random division to generate multiple environmental parameter indication sub-vectors F , F Possibly Kind of value.

[0108] Exemplarily, the target user group includes user 1, user 2, and user 3. The typical environmental parameter set of user 1 is [economic development level parameter, resident income parameter], the typical environmental parameter set of user 2 is [electricity price parameter, temperature parameter], and the typical environmental parameter set of user 3 is [precipitation parameter, wind speed parameter, population density parameter]. The typical environmental parameter sets of user 1, user 2, and user 3 are integrated to obtain the environmental parameter indicator vector of the target user group [economic development level parameter, resident income parameter, electricity price parameter, temperature parameter, precipitation parameter, wind speed parameter, population density parameter]. The number of environmental parameter indicator subvectors obtained after the second random division is random, the number of environmental parameters in the environmental parameter indicator subvector is random, and the type of environmental parameters in the environmental parameter indicator subvector is also random, such as environmental parameter indicator subvector 1 [economic development level parameter, resident income parameter]...environmental parameter indicator subvector 11 [economic development level parameter, resident income parameter, electricity price parameter]...environmental parameter indicator subvector 21 [economic development level parameter, resident income parameter, electricity price parameter, temperature parameter], etc.

[0109] 107. Calculate the feature screening traversal solution index of each environmental parameter indicator subvector respectively, select the maximum feature screening traversal solution index from each feature screening traversal solution index, determine the environmental parameter indicator subvector corresponding to the maximum feature screening traversal solution index as the optimal environmental parameter set of the target user group, and obtain the optimal environmental parameter set of each user group.

[0110] Among them, the feature screening traversal solution index is used to measure the degree of adaptation between the environmental parameter indicator sub-vector and the energy consumption behavior data of the user group.

[0111] Specifically, first, based on the following formula, the characteristic screening traversal solution index of each environmental parameter indicator vector is calculated respectively:

[0112] ,

[0113] in, I Indicates the feature screening traversal solution index, m represents the number of environmental parameters contained in the environmental parameter indicator vector, Indicates environmental parameters and environmental parameters The maximum information coefficient between Indicates environmental parameters The existence mark, Indicates environmental parameters The existence identifier.

[0114] It should be noted that the environmental parameter indicator vector Included m Environmental parameters, after the second random division, can be obtained The environmental parameter indicator subvector F, therefore, Feature screening traversal solution index I .

[0115] Further, from The maximum feature screening traversal solution index is selected from the feature screening traversal solution indexes, and the environmental parameter indicator subvector corresponding to the maximum feature screening traversal solution index is determined as the optimal environmental parameter set for the target user group. , thus obtaining the optimal set of environmental parameters for each user group.

[0116] As an optional implementation method, the maximum objective function of the feature screening traversal solution index can be constructed to screen out the environmental parameter indicator subvector corresponding to the maximum feature screening traversal solution index based on the maximum objective function. The maximum objective function is expressed as the following formula

[0117] ,

[0118] Wherein, F represents the environmental parameter indicator subvector.

[0119] 108. Invoke energy usage behavior data prediction models that match each user group respectively, perform prediction processing based on the optimal environmental parameter set and energy usage behavior data of each user group, and obtain predicted energy usage behavior data of each user group.

[0120] Among them, the energy consumption behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model. The encoder and decoder are built based on long-term and short-term neural networks.

[0121] In an embodiment of the present application, a corresponding energy consumption behavior data prediction model is constructed for each user group, and the model is trained based on the historical data of the user group, so that the model can focus more on the relationship between the environmental parameters of the group and the energy consumption behavior data, thereby improving the accuracy of the model prediction; and the energy consumption behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model, wherein the encoder and the decoder are constructed based on long-short-term neural networks, utilizing the characteristics of long-short-term neural networks with relatively long short-term memory, fully representing the changes in time series, and utilizing the global attention sub-model to consider all hidden layers and learn the weights of different environmental parameters, thereby improving the prediction model's attention to important environmental parameters and reducing the probability of errors gradually accumulating with iterations, thereby improving the accuracy of the energy consumption behavior data prediction results.

[0122] Compared with the prior art, the embodiment of the present application selects a typical environmental parameter set corresponding to each user from the environmental parameter set based on the maximum correlation minimum redundancy index, and then obtains a correlation coefficient vector of each user based on the correlation coefficient combination between each environmental parameter contained in the typical environmental parameter set and the energy usage behavior data of the corresponding user, divides all users into multiple user groups based on the correlation coefficient vector, and then integrates the typical environmental parameter sets of each user contained in the user group to obtain the environmental parameter indication vector of the user group, and determines the optimal environmental parameter set of the user group from the environmental parameter indication vector by using the traversal solution method, which fully considers the correlation between the environmental parameters and the user's energy usage behavior data and the redundancy between the environmental parameters, and is not limited to the user's energy usage behavior data. In order to focus on the data itself, the impact of nonlinear problems such as environmental parameters on user energy consumption behavior data is considered, making the analysis results of user energy consumption characteristics more reliable; further, the energy consumption behavior data prediction model constructed for each user group is composed of an encoder, a decoder, and a global attention sub-model, among which the encoder and decoder are constructed based on long-short-term neural networks, utilizing the characteristics of long-short-term neural networks with longer short-term memory to fully represent the changes in time series, and utilizing the global attention sub-model to consider all hidden layers and learn the weights of different environmental parameters, thereby improving the prediction model's attention to important environmental parameters and reducing the probability of errors gradually accumulating with iterations, thereby improving the accuracy of energy consumption behavior data prediction results.

[0123] In an embodiment of the present application, in order to further define and illustrate, Figure 2 As shown, in step 103 of the embodiment, the maximum correlation minimum redundancy index between the environmental parameter subset and the energy consumption behavior data of the target user is calculated, including:

[0124] 201. Taking each environmental parameter included in the environmental parameter set as a target environmental parameter one by one, and taking each user included in the target area as a target user one by one.

[0125] Among them, the environmental parameter set can be recorded as .

[0126] 202. Based on the following formula, calculate the correlation coefficient between the target environmental parameters and the energy consumption behavior data of the target user:

[0127] ,

[0128] in, Indicates environmental parameters Energy usage behavior data of users The covariance between Indicates environmental parameters The standard deviation of Represents the user's energy usage behavior data The standard deviation of .

[0129] In the embodiment steps 201-202, for example, the environmental parameter set includes environmental parameter a and environmental parameter b, the target area includes user A and user B, and the correlation coefficient between environmental parameter a and user A's energy consumption behavior data is calculated one by one. , the correlation coefficient between environmental parameter a and user B's energy consumption behavior data , the correlation coefficient between environmental parameter b and user A's energy consumption behavior data , the correlation coefficient between environmental parameter b and user B's energy consumption behavior data .

[0130] 203. Obtain a correlation coefficient between each environmental parameter included in the environmental parameter subset and the power usage behavior data of the target user.

[0131] In the embodiment of the present application, the correlation coefficient can be obtained from step 202 of the embodiment.

[0132] 204. Based on the following formula, the maximum correlation index between the environmental parameter subset and the target user's power consumption behavior data is calculated according to each correlation coefficient:

[0133] ,

[0134] in, Represents a subset of environmental parameters Data on electricity usage behavior of users The maximum correlation index between Represents a subset of environmental parameters The number of environment parameters included, Indicates environmental parameters Energy usage behavior data of users The correlation coefficient between .

[0135] 205. Take any two environmental parameters in the environmental parameter subset as target environmental parameter pairs one by one.

[0136] In the embodiment of the present application, the environmental parameters included in the environmental parameter subset are combined in pairs to obtain multiple environmental parameter pairs, and each environmental parameter pair is used as a target environmental parameter pair, that is, a processing object in subsequent steps.

[0137] 206. Disperse the data points included in the two environmental parameters in the target environmental parameter pair in pairs in a two-dimensional space to obtain a data point set of the target environmental parameter pair.

[0138] In the embodiment of the present application, the environmental parameters are all environmental parameter sequences, that is, sequences containing multiple data points. The two environmental parameters in the target environmental parameter pair can be respectively recorded as as well as , the environmental parameters And environmental parameters The data points contained are scattered in pairs in two-dimensional space, and the data point set of the target environmental parameter pair can be obtained. .

[0139] 207. According to different division scales, the two-dimensional space is gridded to obtain multiple two-dimensional space grids, and the mutual information index of the data point set in each two-dimensional space grid is calculated based on the following formula:

[0140] ,

[0141] in, Represents a collection of data points In two-dimensional space The mutual information indicator in represents the environmental parameters, A represents the interval in the x-axis direction of the two-dimensional space, and B represents the interval in the y-axis direction of the two-dimensional space. Indicates that the data points in the data point set are located in the region The probability of Indicates that the data points in the data point set are located in the region The probability of Indicates that the data points in the data point set are located in the region The probability of express from Take the value in express from Take the value from.

[0142] 208. Select a maximum mutual information index from the multiple mutual information indexes, and normalize the maximum mutual information index based on the following formula to obtain the normalized maximum mutual information index.

[0143] ,

[0144] in, Represents the maximum mutual information index of the data point set after normalization in the two-dimensional space grid, Represents the maximum mutual information index of a data point set in a two-dimensional space grid, Represents the number of grids in the two-dimensional space grid in the x-axis direction. Indicates the number of grids in the y-axis direction of the two-dimensional space grid.

[0145] 209. Based on the following formula, the maximum information coefficient of the target environment parameter pair is calculated according to the maximum mutual information index after normalization:

[0146] ,

[0147] in, Represents the mesh complexity parameter.

[0148] It should be noted that Usually the value of the data point set The number of data points in the grid is 0.6 to control the complexity of the grid, and can be adjusted according to the actual accuracy requirements. The value of .

[0149] 210. Based on the following formula, the minimum redundancy index of the environmental parameter subset is calculated according to the maximum information coefficient of each target environmental parameter pair:

[0150] ,

[0151] in, Represents the minimum redundancy index of a subset of environmental parameters.

[0152] 211. The difference between the maximum correlation index and the minimum redundancy index is determined as the maximum correlation minimum redundancy index between the environmental parameter subset and the energy consumption behavior data of the target user.

[0153] Furthermore, as an optional implementation method, a maximum objective function of the maximum relevant minimum redundancy index can be constructed to screen out a subset of environmental parameters corresponding to the maximum value from multiple maximum relevant minimum redundancy indicators corresponding to the target user based on the maximum objective function. The maximum objective function is expressed as the following formula:

[0154] ,

[0155] in, represents the maximum related minimum redundancy index, Represents a typical set of environmental parameters.

[0156] Based on the above maximum objective function, we solve the maximum correlation and minimum redundancy index The subset of environmental parameters that obtains the maximum value is the typical environmental parameter set. .

[0157] In an embodiment of the present application, in order to further define and illustrate, Figure 3 As shown, before respectively calling the energy consumption behavior data prediction models matching each user group in step 108 of the embodiment, the embodiment method further includes:

[0158] 301. Construct a corresponding initial energy usage behavior data prediction model for each user group, and use each initial energy usage behavior data prediction model as a target initial energy usage behavior data prediction model.

[0159] Among them, the initial energy consumption behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model. The encoder and decoder are built based on long-term and short-term neural networks.

[0160] First, the encoder and decoder are constructed based on the long-short term neural network. The structure of the long-short term neural network is as follows:

[0161] ,

[0162] ,

[0163] ,

[0164] ,

[0165] ,

[0166] ,

[0167] in, represents the input gate, represents the forget gate, represents the output gate, represents the hidden state of the current unit, Represents a new input signal, The signal representing the input at the previous moment, represents the output vector, represents the output vector at the previous moment, represents the input vector of LSTM at the current moment, sigm represents the activation function, tanh represents the hyperbolic tangent function, , , , Both represent learnable weights, , , , Both represent learnable biases.

[0168] Furthermore, an initial energy consumption behavior data prediction model is constructed based on the S2S framework, such as Figure 4 As shown, the model is composed of the encoder and decoder established above. The optimal environmental parameter set and energy consumption behavior data of the user group are input from the encoder side, and the encoder encodes them into a fixed-length vector, which is input into the decoder as the input set of the decoder, and the decoder generates an output vector.

[0169] Furthermore, in order to increase the attention to important environmental parameters and reduce the probability of gradual accumulation of errors with iterations, in an embodiment of the present application, a global attention sub-model is added after the decoder output vector to calculate the similarity between the hidden state of the decoder in the current stage and the hidden state of the encoder in each iterative stage, and jointly determine the output result with the encoder hidden state to improve the accuracy of the energy consumption behavior data prediction results.

[0170] In the embodiment of the present application, a corresponding energy consumption behavior data prediction model is constructed for each user group, so that the model can focus more on the relationship between the environmental parameters and energy consumption behavior data of the group, thereby improving the accuracy of the model prediction; and the long-term and short-term neural network has the characteristics of relatively long short-term memory, which fully represents the changes in the time series, and uses the global attention sub-model to consider all hidden layers and learn the weights of different environmental parameters, thereby improving the prediction model's attention to important environmental parameters and reducing the probability of errors gradually accumulating with iterations, thereby improving the accuracy of the energy consumption behavior data prediction results.

[0171] 302. Obtain energy usage behavior data of a user group corresponding to a target initial energy usage behavior data prediction model in a first historical period and a historical optimal environmental parameter set.

[0172] Among them, the duration of the first historical period may be the same as or different from the preset duration in step 101 of the embodiment, and is not specifically limited in the embodiment of the present application; the process of determining the historical optimal environmental parameter set is the same as the process of determining the above-mentioned optimal environmental parameter set, and will not be repeated here.

[0173] 303. Input the energy consumption behavior data of the first historical period and the historical optimal environmental parameter set into the target initial energy consumption behavior data prediction model to obtain the predicted energy consumption behavior data of the second historical period.

[0174] Among them, the natural time sequence of the second historical period is after the natural time sequence of the first historical period, that is, the first historical period occurs before the second historical period.

[0175] 304. Perform model training on the target initial energy consumption behavior data prediction model based on the loss function between the predicted energy consumption behavior data and the actual energy consumption behavior data in the second historical period to obtain the target energy consumption behavior data prediction model for which the model training has been completed.

[0176] Among them, the loss function can use mean square error, mean absolute error, cross entropy loss, etc.

[0177] 305. Obtain energy usage behavior data prediction models corresponding to each user group that have completed model training, so as to predict the predicted energy usage behavior data of each user group based on the energy usage behavior data prediction models that have completed model training.

[0178] In a specific application scenario, the energy consumption behavior data and the optimal environmental parameter set are input into the encoder, which converts them into a fixed-length vector and then inputs the vector into the hidden layer to generate the encoder hidden state and cell state, and generate the input vector encoding. Furthermore, the input vector encoding is input into the decoder, and the decoder hidden state and cell state are calculated based on the input vector encoding and the state at the previous moment. Furthermore, before the decoder outputs the prediction result, the global attention sub-model is used to calculate the similarity between the decoder hidden state and the encoder hidden state, and the output result is determined together with the encoder hidden state to obtain the predicted energy consumption behavior data. Specifically, the hidden state generated by the encoder in the last time step is passed through the weight matrix Generate a scalar as the starting information, which is the known information at the prediction moment of the decoder's first time step input are concatenated and used together as the input of the decoder at the first time step. For example, Figure 5 As shown, the decoder hidden state generated in the current iteration stage is calculated consistent with the encoder hidden state to obtain a global consistency weight vector, whose length is equal to the number of encoder iterations, which can be calculated based on the following formula:

[0179] ,

[0180] in, represents the global consistency weight vector, softm represents the activation function, represents the hidden state of the decoder at the current iteration stage, represents the hidden state of the encoder, Represents the hidden state of the decoder at the current iteration stage and the encoder’s hidden state The dot product between Indicates that the encoder is The hidden state at the iteration stage, represents the number of hidden layer states, and exp represents the exponential function. Further, the global consistency weight vector Hidden state for encoder Weighted summation to obtain the semantic vector ; Then the hidden state of the decoder at the current iteration stage and semantic vector Obtaining attention hidden states through concatenation , then the target output of the decoder at the current stage is expressed as the following formula:

[0181] ,

[0182] in, represents the target output of the decoder at the current stage, represents the weight matrix of the decoder, Indicates the hidden state of attention. Then the known information The target output of the current decoder stage The concatenation is iterated as input to the next iterative stage of the decoder.

[0183] In an embodiment of the present application, in order to further define and illustrate, Figure 6 As shown, in step 105 of the embodiment, all users are clustered based on the correlation coefficient matrix to obtain multiple user groups, including:

[0184] 401. Based on the number of pre-configured user groups, randomly select the number of users as initial cluster centroids.

[0185] Optionally, the number of user groups may be pre-configured using an elbow method. Specifically, the number of multiple user groups may be pre-configured, which may be configured based on experience.

[0186] The sum of squared errors within the cluster is calculated for each number of user groups, and a curve of the sum of squared errors within the cluster varying with the number of user groups is generated. The calculation formula for the sum of squared errors within the cluster is expressed as follows:

[0187] ,

[0188] in, SSE represents the sum of squared errors within the cluster, K Indicates the number of user groups. pIndicates user group Users in Indicates user group The cluster centroid of

[0189] Based on the elbow method, the optimal number of user groups is determined according to the curve, so as to select the initial cluster centroids based on the optimal number of user groups.

[0190] Among them, based on the elbow method, the optimal number of user groups is determined according to the curve, that is, the inflection point where the downward trend turns to a flat trend is found in the curve of the sum of squared errors within the cluster changing with the number of user groups. This inflection point is the optimal number of user groups, and the initial cluster centroid is selected based on the optimal number of user groups.

[0191] 402. Calculate the Euclidean distance between the correlation coefficient vector of each user and the correlation coefficient vector of each user corresponding to the initial clustering centroid, select the initial clustering centroid corresponding to the minimum Euclidean distance for each user, and assign each user to the cluster corresponding to the initial clustering centroid.

[0192] Among them, the correlation coefficient vector is obtained from the correlation coefficient matrix. The Euclidean distance calculation formula is expressed as the following formula

[0193] ,

[0194] in, Indicates user The correlation coefficient vector Correlation coefficient vector of the user corresponding to the initial cluster centroid The Euclidean distance between Indicates environmental parameters Energy usage behavior data of users The correlation coefficient between Indicates environmental parameters Energy usage behavior data of users The correlation coefficient between m Represents the number of environmental parameters included in the correlation coefficient vector.

[0195] 403. Calculate the center vector of each cluster, and determine the center vector as the updated cluster centroid.

[0196] In the embodiment of the present application, the calculation formula of the center vector is expressed as the following formula

[0197] ,

[0198] in, Represents clusters The number of users included in Indicates updating cluster centroids.

[0199] Furthermore, after obtaining each initial clustering cluster, the cluster centroid of each initial clustering cluster is updated respectively.

[0200] 404. Calculate the Euclidean distance between the correlation coefficient vector of each user and each updated cluster centroid respectively, and reallocate each user to the cluster after the cluster centroid is updated, until a preset number of iterations is reached or the center vector of each cluster remains unchanged, thereby obtaining multiple user groups.

[0201] In the embodiment of the present application, the contents of steps 402-403 of the embodiment are repeated to update the clustering clusters until a preset number of iterations is reached or the center vectors of each clustering cluster remain unchanged, and the multiple updated clustering clusters obtained are the user groups that have completed clustering.

[0202] The present application provides an energy consumption data prediction method based on feature screening. Compared with the prior art, the embodiment of the present application screens out a typical environmental parameter set corresponding to each user from the environmental parameter set based on the maximum correlation minimum redundancy index, and then obtains the correlation coefficient vector of each user based on the correlation coefficient combination between each environmental parameter contained in the typical environmental parameter set and the energy consumption behavior data of the corresponding user, divides all users into multiple user groups based on the correlation coefficient vector, and then integrates the typical environmental parameter sets of each user contained in the user group to obtain the environmental parameter indication vector of the user group, and uses the traversal solution method to determine the optimal environmental parameter set of the user group from the environmental parameter indication vector, which fully considers the correlation between the environmental parameters and the user's energy consumption behavior data and the redundancy between the environmental parameters. , is not limited to the attention on the user energy consumption behavior data itself, but takes into account the impact of nonlinear problems such as environmental parameters on the user energy consumption behavior data, making the analysis results of user energy consumption characteristics more reliable; further, the energy consumption behavior data prediction model constructed for each user group is composed of an encoder, a decoder, and a global attention sub-model, wherein the encoder and the decoder are constructed based on long-short-term neural networks, which utilize the characteristics of long-short-term neural networks with relatively long short-term memory, fully represent the changes in time series, and use the global attention sub-model to consider all hidden layers and learn the weights of different environmental parameters, thereby improving the prediction model's attention to important environmental parameters and reducing the probability of errors gradually accumulating with iterations, thereby improving the accuracy of energy consumption behavior data prediction results.

[0203] Furthermore, as a response to the above Figure 1 The implementation of the method shown in the embodiment of the present application provides an energy consumption data prediction device based on feature screening, such as Figure 7 As shown, the device comprises:

[0204] Data acquisition module 51, environment parameter subset division module 52, typical environment parameter set determination module 53, correlation coefficient matrix determination module 54, user clustering module 55, environment parameter indication subvector division module 56, optimal environment parameter set determination module 57, energy use behavior data prediction module 58;

[0205] A data acquisition module 51 is used to acquire energy usage behavior data of all users included in a target area within a preset time period and an environmental parameter set of the target area;

[0206] An environmental parameter subset division module 52 is used to perform a first random division on the multiple environmental parameters included in the environmental parameter set to generate multiple environmental parameter subsets, wherein the first random division includes the number of environmental parameter subsets, the number of environmental parameters in the environmental parameter subsets, and the types of environmental parameters in the environmental parameter subsets;

[0207] The typical environment parameter set determination module 53 is used to take each of the users as a target user one by one, and respectively calculate the maximum correlation minimum redundancy index between each of the environment parameter subsets and the energy consumption behavior data of the target user, and select the environment parameter subset corresponding to the maximum value from the multiple maximum correlation minimum redundancy indexes corresponding to the target user, and take the environment parameter subset as the typical environment parameter set of the target user, and the maximum correlation minimum redundancy index is used to characterize the difference between the maximum correlation index and the minimum redundancy index between the environment parameter subset and the energy consumption behavior data of the user;

[0208] A correlation coefficient matrix determination module 54 is used to respectively calculate the correlation coefficient between each environmental parameter included in the typical environmental parameter set and the energy usage behavior data of the target user, combine each of the correlation coefficients to obtain a correlation coefficient vector of the target user, and combine each of the correlation coefficient vectors of the target user to obtain a correlation coefficient matrix between the user and the typical environmental parameter set;

[0209] A user clustering module 55 is used to perform clustering processing on all users based on the correlation coefficient matrix to obtain multiple user groups, and use each of the user groups as a target user group;

[0210] An environment parameter indication sub-vector division module 56 is used to integrate the typical environment parameter sets of each user included in the target user group to obtain the environment parameter indication vector of the target user group, and perform a second random division on the multiple environment parameters included in the environment parameter indication vector to generate multiple environment parameter indication sub-vectors, wherein the second random division includes the number of environment parameter indication sub-vectors, the number of environment parameters in the environment parameter indication sub-vectors, and the types of environment parameters in the environment parameter indication sub-vectors;

[0211] The optimal environmental parameter set determination module 57 is used to respectively calculate the characteristic screening traversal solution index of each of the environmental parameter indication sub-vectors, screen out the maximum characteristic screening traversal solution index from each of the characteristic screening traversal solution indexes, determine the environmental parameter indication sub-vector corresponding to the maximum characteristic screening traversal solution index as the optimal environmental parameter set of the target user group, and obtain the optimal environmental parameter set of each user group, wherein the characteristic screening traversal solution index is used to measure the degree of adaptability between the environmental parameter indication sub-vector and the energy consumption behavior data of the user group;

[0212] The energy usage behavior data prediction module 58 is used to call the energy usage behavior data prediction model matching each of the user groups respectively, perform prediction processing according to the optimal environmental parameter set and energy usage behavior data of each of the user groups, and obtain the predicted energy usage behavior data of each of the user groups. The energy usage behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model, and the encoder and the decoder are constructed based on long-short-term neural networks.

[0213] In a specific application scenario, the typical environmental parameter set determination module is used to:

[0214] Obtaining a correlation coefficient between each environmental parameter included in the environmental parameter subset and the power consumption behavior data of the target user;

[0215] Based on the following formula, the maximum correlation index between the environmental parameter subset and the power consumption behavior data of the target user is calculated according to each of the correlation coefficients:

[0216] ,

[0217] in, Represents a subset of environmental parameters Data on electricity usage behavior of users The maximum correlation index between Represents a subset of environmental parameters The number of environment parameters included, Indicates environmental parameters Energy usage behavior data of users The correlation coefficient between

[0218] Taking any two environmental parameters in the environmental parameter subset as target environmental parameter pairs one by one;

[0219] Dispersing the data points included in the two environmental parameters in the target environmental parameter pair in pairs in a two-dimensional space to obtain a data point set of the target environmental parameter pair;

[0220] According to different division scales, the two-dimensional space is gridded to obtain multiple two-dimensional space grids, and the mutual information index of the data point set in each of the two-dimensional space grids is calculated based on the following formula:

[0221] ,

[0222] in, Represents a collection of data points In two-dimensional space The mutual information indicator in represents the environmental parameters, A represents the interval in the x-axis direction of the two-dimensional space, and B represents the interval in the y-axis direction of the two-dimensional space. Indicates that the data points in the data point set are located in the region The probability of Indicates that the data points in the data point set are located in the region The probability of Indicates that the data points in the data point set are located in the region The probability of express from Take the value in express from Take the value in

[0223] The maximum mutual information index is selected from the plurality of mutual information indexes, and the maximum mutual information index is normalized based on the following formula to obtain the normalized maximum mutual information index.

[0224] ,

[0225] in, Represents the maximum mutual information index of the data point set after normalization in the two-dimensional space grid, Represents the maximum mutual information index of a data point set in a two-dimensional space grid, Represents the number of grids in the two-dimensional space grid in the x-axis direction. Represents the number of grids in the y-axis direction of the two-dimensional space grid;

[0226] Based on the following formula, the maximum information coefficient of the target environment parameter pair is calculated according to the maximum mutual information index after the normalization process:

[0227] ,

[0228] in, Indicates environmental parameters and environmental parameters The maximum information coefficient between represents the mesh complexity parameter;

[0229] Based on the following formula, the minimum redundancy index of the environmental parameter subset is calculated according to the maximum information coefficient of each target environmental parameter pair:

[0230] ,

[0231] in, represents the minimum redundancy index of a subset of environmental parameters;

[0232] The difference between the maximum correlation index and the minimum redundancy index is determined as the maximum correlation minimum redundancy index between the environmental parameter subset and the energy usage behavior data of the target user.

[0233] In a specific application scenario, the typical environment parameter set determination module is also used to:

[0234] Taking each environmental parameter included in the environmental parameter set as a target environmental parameter one by one, and taking each user included in the target area as a target user one by one;

[0235] Based on the following formula, the correlation coefficient between the target environmental parameters and the target user's energy consumption behavior data is calculated.

[0236] ,

[0237] in, Indicates environmental parameters Energy usage behavior data of users The covariance between Indicates environmental parameters The standard deviation of Represents the user's energy usage behavior data The standard deviation of .

[0238] In a specific application scenario, before the energy consumption behavior data prediction module, the device further includes a model training module for:

[0239] Constructing a corresponding initial energy consumption behavior data prediction model for each of the user groups, and using each of the initial energy consumption behavior data prediction models as a target initial energy consumption behavior data prediction model, wherein the initial energy consumption behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model, and the encoder and the decoder are constructed based on a long-short term neural network;

[0240] Obtaining energy usage behavior data of a user group corresponding to the target initial energy usage behavior data prediction model in a first historical period and a historical optimal environmental parameter set;

[0241] Inputting the energy usage behavior data of the first historical period and the historical optimal environmental parameter set into the target initial energy usage behavior data prediction model to obtain predicted energy usage behavior data of the second historical period;

[0242] Performing model training on the target initial energy usage behavior data prediction model based on a loss function between the predicted energy usage behavior data and the actual energy usage behavior data of the second historical period to obtain a target energy usage behavior data prediction model for which model training has been completed;

[0243] The energy usage behavior data prediction models corresponding to the respective user groups that have completed model training are obtained, so as to predict the predicted energy usage behavior data of the respective user groups based on the energy usage behavior data prediction models that have completed model training.

[0244] In a specific application scenario, the user clustering module is used to:

[0245] Based on the number of pre-configured user groups, randomly select the number of users as initial cluster centroids;

[0246] Calculating the Euclidean distance between the correlation coefficient vector of each user and the correlation coefficient vector of each user corresponding to the initial clustering centroid, selecting the initial clustering centroid corresponding to the minimum Euclidean distance for each user, and assigning each user to the cluster corresponding to the initial clustering centroid, wherein the correlation coefficient vector is obtained from the correlation coefficient matrix;

[0247] Calculating the center vector of each cluster, and determining the center vector as the updated cluster centroid;

[0248] The Euclidean distance between the correlation coefficient vector of each user and each updated cluster centroid is calculated respectively, and each user is reallocated to the cluster after the cluster centroid is updated, until a preset number of iterations is reached or the center vector of each cluster remains unchanged, thereby obtaining multiple user groups.

[0249] In a specific application scenario, the user clustering module is also used to:

[0250] Pre-configure multiple user groups;

[0251] Calculate the sum of squared errors within each cluster under different numbers of user groups, and generate a curve showing the sum of squared errors within the cluster changing with the number of user groups;

[0252] Based on the elbow method, the optimal number of user groups is determined according to the curve, so as to select initial cluster centroids based on the optimal number of user groups.

[0253] In a specific application scenario, the environmental parameter set includes at least economic development level parameters, resident income parameters, electricity price parameters, temperature parameters, precipitation parameters, wind speed parameters, population density parameters, and date type parameters.

[0254] The present application provides an energy consumption data prediction device based on feature screening. Compared with the prior art, the embodiment of the present application screens out a typical environmental parameter set corresponding to each user from the environmental parameter set based on the maximum correlation minimum redundancy index, and then obtains the correlation coefficient vector of each user based on the correlation coefficient combination between each environmental parameter contained in the typical environmental parameter set and the energy consumption behavior data of the corresponding user, divides all users into multiple user groups based on the correlation coefficient vector, and then integrates the typical environmental parameter sets of each user contained in the user group to obtain the environmental parameter indication vector of the user group, and uses the traversal solution method to determine the optimal environmental parameter set of the user group from the environmental parameter indication vector, which fully considers the correlation between the environmental parameters and the user's energy consumption behavior data and the redundancy between the environmental parameters. , is not limited to the attention on the user energy consumption behavior data itself, but takes into account the impact of nonlinear problems such as environmental parameters on the user energy consumption behavior data, making the analysis results of user energy consumption characteristics more reliable; further, the energy consumption behavior data prediction model constructed for each user group is composed of an encoder, a decoder, and a global attention sub-model, wherein the encoder and the decoder are constructed based on long-short-term neural networks, which utilize the characteristics of long-short-term neural networks with relatively long short-term memory, fully represent the changes in time series, and use the global attention sub-model to consider all hidden layers and learn the weights of different environmental parameters, thereby improving the prediction model's attention to important environmental parameters and reducing the probability of errors gradually accumulating with iterations, thereby improving the accuracy of energy consumption behavior data prediction results.

[0255] According to an embodiment of the present application, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the computer executable instruction can execute the energy consumption data prediction method based on feature screening in any of the above method embodiments.

[0256] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0257] Figure 8 A schematic diagram of the structure of a terminal provided according to an embodiment of the present application is shown, and the specific embodiment of the present application does not limit the specific implementation of the terminal.

[0258] like Figure 8 As shown, the terminal may include: a processor (processor) 602 , a communication interface (Communications Interface) 604 , a memory (memory) 606 , and a communication bus 608 .

[0259] The processor 602 , the communication interface 604 , and the memory 606 communicate with each other via a communication bus 608 .

[0260] The communication interface 604 is used to communicate with other devices such as clients or other servers.

[0261] The processor 602 is used to execute the program 610, and specifically can execute the relevant steps in the above-mentioned embodiment of the energy consumption data prediction method based on feature screening.

[0262] Specifically, the program 610 may include program codes, which include computer operation instructions.

[0263] The processor 602 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the computer device may be processors of the same type, such as one or more CPUs; or may be processors of different types, such as one or more CPUs and one or more ASICs.

[0264] The memory 606 is used to store the program 610. The memory 606 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0265] The program 610 may be specifically configured to enable the processor 602 to perform the following operations:

[0266] Obtaining energy usage behavior data of all users in the target area within a preset time period and an environmental parameter set of the target area;

[0267] Performing a first random division on the multiple environmental parameters included in the environmental parameter set to generate multiple environmental parameter subsets, wherein the first random division includes the number of environmental parameter subsets, the number of environmental parameters in the environmental parameter subsets, and the types of environmental parameters in the environmental parameter subsets;

[0268] Each of the users is taken as a target user one by one, and the maximum correlation minimum redundancy index between each of the environmental parameter subsets and the energy consumption behavior data of the target user is calculated respectively, and the environmental parameter subset corresponding to the maximum value is screened out from the multiple maximum correlation minimum redundancy indexes corresponding to the target user, and the environmental parameter subset is taken as the typical environmental parameter set of the target user, and the maximum correlation minimum redundancy index is used to characterize the difference between the maximum correlation index and the minimum redundancy index between the environmental parameter subset and the energy consumption behavior data of the user;

[0269] Calculating the correlation coefficient between each environmental parameter included in the typical environmental parameter set and the energy usage behavior data of the target user respectively, combining each of the correlation coefficients to obtain a correlation coefficient vector of the target user, and combining each of the correlation coefficient vectors of the target user to obtain a correlation coefficient matrix between the user and the typical environmental parameter set;

[0270] Performing clustering processing on all users based on the correlation coefficient matrix to obtain multiple user groups, and taking each of the user groups as a target user group;

[0271] Integrate the typical environment parameter sets of each user included in the target user group to obtain an environment parameter indication vector of the target user group, perform a second random division on multiple environment parameters included in the environment parameter indication vector to generate multiple environment parameter indication sub-vectors, wherein the second random division includes the number of environment parameter indication sub-vectors, the number of environment parameters in the environment parameter indication sub-vectors, and the types of environment parameters in the environment parameter indication sub-vectors;

[0272] Calculate the characteristic screening traversal solution index of each of the environmental parameter indication sub-vectors respectively, select the maximum characteristic screening traversal solution index from each of the characteristic screening traversal solution indexes, determine the environmental parameter indication sub-vector corresponding to the maximum characteristic screening traversal solution index as the optimal environmental parameter set of the target user group, and obtain the optimal environmental parameter set of each user group, wherein the characteristic screening traversal solution index is used to measure the degree of adaptability between the environmental parameter indication sub-vector and the energy consumption behavior data of the user group;

[0273] The energy usage behavior data prediction models matching each of the user groups are called respectively, and prediction processing is performed according to the optimal environmental parameter set and energy usage behavior data of each of the user groups to obtain the predicted energy usage behavior data of each of the user groups. The energy usage behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model, and the encoder and the decoder are constructed based on long-short term neural networks.

[0274] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the physical device of the above-mentioned energy consumption data prediction method based on feature screening, and supports the operation of information processing programs and other software and / or programs. The network communication module is used to realize communication between components inside the storage medium, and communication with other hardware and software in the information processing physical device.

[0275] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting energy consumption data based on feature screening, characterized in that: include: Obtaining energy usage behavior data of all users in the target area within a preset time period and an environmental parameter set of the target area; Performing a first random division on the multiple environmental parameters included in the environmental parameter set to generate multiple environmental parameter subsets, wherein the first random division includes the number of environmental parameter subsets, the number of environmental parameters in the environmental parameter subsets, and the types of environmental parameters in the environmental parameter subsets; Each of the users is taken as a target user one by one, and the maximum correlation minimum redundancy index between each of the environmental parameter subsets and the energy consumption behavior data of the target user is calculated respectively, and the environmental parameter subset corresponding to the maximum value is screened out from the multiple maximum correlation minimum redundancy indexes corresponding to the target user, and the environmental parameter subset is taken as the typical environmental parameter set of the target user, and the maximum correlation minimum redundancy index is used to characterize the difference between the maximum correlation index and the minimum redundancy index between the environmental parameter subset and the energy consumption behavior data of the user; Calculating the correlation coefficient between each environmental parameter included in the typical environmental parameter set and the energy usage behavior data of the target user respectively, combining each of the correlation coefficients to obtain a correlation coefficient vector of the target user, and combining each of the correlation coefficient vectors of the target user to obtain a correlation coefficient matrix between the user and the typical environmental parameter set; Performing clustering processing on all users based on the correlation coefficient matrix to obtain multiple user groups, and taking each of the user groups as a target user group; Integrate the typical environment parameter sets of each user included in the target user group to obtain an environment parameter indication vector of the target user group, perform a second random division on multiple environment parameters included in the environment parameter indication vector to generate multiple environment parameter indication sub-vectors, wherein the second random division includes the number of environment parameter indication sub-vectors, the number of environment parameters in the environment parameter indication sub-vectors, and the types of environment parameters in the environment parameter indication sub-vectors; Calculate the characteristic screening traversal solution index of each of the environmental parameter indication sub-vectors respectively, select the maximum characteristic screening traversal solution index from each of the characteristic screening traversal solution indexes, determine the environmental parameter indication sub-vector corresponding to the maximum characteristic screening traversal solution index as the optimal environmental parameter set of the target user group, and obtain the optimal environmental parameter set of each user group, wherein the characteristic screening traversal solution index is used to measure the degree of adaptability between the environmental parameter indication sub-vector and the energy consumption behavior data of the user group; The energy usage behavior data prediction models matching each of the user groups are called respectively, and prediction processing is performed according to the optimal environmental parameter set and energy usage behavior data of each of the user groups to obtain the predicted energy usage behavior data of each of the user groups. The energy usage behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model, and the encoder and the decoder are constructed based on long-short term neural networks.

2. The method according to claim 1, characterized in that The calculating of the maximum correlation minimum redundancy index between the environmental parameter subset and the energy usage behavior data of the target user includes: Obtaining a correlation coefficient between each environmental parameter included in the environmental parameter subset and the power consumption behavior data of the target user; Based on the following formula, the maximum correlation index between the environmental parameter subset and the power consumption behavior data of the target user is calculated according to each of the correlation coefficients: ， in, Represents a subset of environmental parameters Data on electricity usage behavior of users The maximum correlation index between Represents a subset of environmental parameters The number of environment parameters included, Indicates environmental parameters Energy usage behavior data of users The correlation coefficient between Taking any two environmental parameters in the environmental parameter subset as target environmental parameter pairs one by one; Dispersing the data points included in the two environmental parameters in the target environmental parameter pair in pairs in a two-dimensional space to obtain a data point set of the target environmental parameter pair; According to different division scales, the two-dimensional space is gridded to obtain multiple two-dimensional space grids, and the mutual information index of the data point set in each of the two-dimensional space grids is calculated based on the following formula: ， in, Represents a collection of data points In two-dimensional space The mutual information indicator in represents the environmental parameters, A represents the interval in the x-axis direction of the two-dimensional space, and B represents the interval in the y-axis direction of the two-dimensional space. Indicates that the data points in the data point set are located in the region The probability of Indicates that the data points in the data point set are located in the region The probability of Indicates that the data points in the data point set are located in the region The probability of express from Take the value in express from Take the value in The maximum mutual information index is selected from the plurality of mutual information indexes, and the maximum mutual information index is normalized based on the following formula to obtain the normalized maximum mutual information index. ， in, Represents the maximum mutual information index of the data point set after normalization in the two-dimensional space grid, Represents the maximum mutual information index of a data point set in a two-dimensional space grid, Represents the number of grids in the two-dimensional space grid in the x-axis direction. Represents the number of grids in the y-axis direction of the two-dimensional space grid; Based on the following formula, the maximum information coefficient of the target environment parameter pair is calculated according to the maximum mutual information index after the normalization process: ， in, Indicates environmental parameters and environmental parameters The maximum information coefficient between represents the mesh complexity parameter; Based on the following formula, the minimum redundancy index of the environmental parameter subset is calculated according to the maximum information coefficient of each target environmental parameter pair: ， in, represents the minimum redundancy index of a subset of environmental parameters; The difference between the maximum correlation index and the minimum redundancy index is determined as the maximum correlation minimum redundancy index between the environmental parameter subset and the energy usage behavior data of the target user.

3. The method according to claim 2, characterized in that Before obtaining the correlation coefficient between each environmental parameter included in the environmental parameter subset and the power usage behavior data of the target user, the method further includes: Taking each environmental parameter included in the environmental parameter set as a target environmental parameter one by one, and taking each user included in the target area as a target user one by one; Based on the following formula, the correlation coefficient between the target environmental parameters and the target user's energy consumption behavior data is calculated. ， in, Indicates environmental parameters Energy usage behavior data of users The covariance between Indicates environmental parameters The standard deviation of Represents the user's energy usage behavior data The standard deviation of .

4. The method according to claim 1, characterized in that: Before respectively calling the energy consumption behavior data prediction models matching the respective user groups, the method further includes: Constructing a corresponding initial energy consumption behavior data prediction model for each of the user groups, and using each of the initial energy consumption behavior data prediction models as a target initial energy consumption behavior data prediction model, wherein the initial energy consumption behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model, and the encoder and the decoder are constructed based on a long-short term neural network; Obtaining energy usage behavior data of a user group corresponding to the target initial energy usage behavior data prediction model in a first historical period and a historical optimal environmental parameter set; Inputting the energy usage behavior data of the first historical period and the historical optimal environmental parameter set into the target initial energy usage behavior data prediction model to obtain predicted energy usage behavior data of the second historical period; Performing model training on the target initial energy usage behavior data prediction model based on a loss function between the predicted energy usage behavior data and the actual energy usage behavior data of the second historical period to obtain a target energy usage behavior data prediction model for which model training has been completed; The energy usage behavior data prediction models corresponding to the respective user groups that have completed model training are obtained, so as to predict the predicted energy usage behavior data of the respective user groups based on the energy usage behavior data prediction models that have completed model training.

5. The method according to claim 1, characterized in that: The clustering process is performed on all users based on the correlation coefficient matrix to obtain multiple user groups, including: Based on the number of pre-configured user groups, randomly select the number of users as initial cluster centroids; Calculating the Euclidean distance between the correlation coefficient vector of each user and the correlation coefficient vector of each user corresponding to the initial clustering centroid, selecting the initial clustering centroid corresponding to the minimum Euclidean distance for each user, and assigning each user to the cluster corresponding to the initial clustering centroid, wherein the correlation coefficient vector is obtained from the correlation coefficient matrix; Calculating the center vector of each cluster, and determining the center vector as the updated cluster centroid; The Euclidean distance between the correlation coefficient vector of each user and each updated cluster centroid is calculated respectively, and each user is reallocated to the cluster after the cluster centroid is updated, until a preset number of iterations is reached or the center vector of each cluster remains unchanged, thereby obtaining multiple user groups.

6. The method according to claim 5, characterized in that Before randomly selecting the number of users as initial cluster centroids based on the number of pre-configured user groups, the method further includes: Pre-configure multiple user groups; Calculate the sum of squared errors within each cluster under different numbers of user groups, and generate a curve showing the sum of squared errors within the cluster changing with the number of user groups; Based on the elbow method, the optimal number of user groups is determined according to the curve, so as to select initial cluster centroids based on the optimal number of user groups.

7. The method according to claim 1, characterized in that The environmental parameter set includes at least economic development level parameters, resident income parameters, electricity price parameters, temperature parameters, precipitation parameters, wind speed parameters, population density parameters, and date type parameters.

8. An energy consumption data prediction device based on feature screening, characterized in that: include: A data acquisition module, used to acquire energy usage behavior data of all users included in the target area within a preset time period and an environmental parameter set of the target area; An environmental parameter subset division module, used for performing a first random division on a plurality of environmental parameters included in the environmental parameter set to generate a plurality of environmental parameter subsets, wherein the first random division includes the number of environmental parameter subsets, the number of environmental parameters in the environmental parameter subsets, and the types of environmental parameters in the environmental parameter subsets; A typical environmental parameter set determination module, used to take each of the users as a target user one by one, and respectively calculate the maximum correlation minimum redundancy index between each of the environmental parameter subsets and the energy consumption behavior data of the target user, and screen out the environmental parameter subset corresponding to the maximum value from the multiple maximum correlation minimum redundancy indexes corresponding to the target user, and take the environmental parameter subset as the typical environmental parameter set of the target user, wherein the maximum correlation minimum redundancy index is used to characterize the difference between the maximum correlation index and the minimum redundancy index between the environmental parameter subset and the energy consumption behavior data of the user; A correlation coefficient matrix determination module, used to respectively calculate the correlation coefficient between each environmental parameter included in the typical environmental parameter set and the energy usage behavior data of the target user, combine each of the correlation coefficients to obtain a correlation coefficient vector of the target user, and combine each of the correlation coefficient vectors of the target user to obtain a correlation coefficient matrix between the user and the typical environmental parameter set; A user clustering module, used to perform clustering processing on all users based on the correlation coefficient matrix to obtain multiple user groups, and use each of the user groups as a target user group; an environment parameter indication sub-vector division module, used to integrate the typical environment parameter sets of each user included in the target user group to obtain the environment parameter indication vector of the target user group, and perform a second random division on multiple environment parameters included in the environment parameter indication vector to generate multiple environment parameter indication sub-vectors, wherein the second random division includes the number of environment parameter indication sub-vectors, the number of environment parameters in the environment parameter indication sub-vectors, and the types of environment parameters in the environment parameter indication sub-vectors; An optimal environmental parameter set determination module is used to respectively calculate the characteristic screening traversal solution index of each of the environmental parameter indication sub-vectors, screen out the maximum characteristic screening traversal solution index from each of the characteristic screening traversal solution indexes, determine the environmental parameter indication sub-vector corresponding to the maximum characteristic screening traversal solution index as the optimal environmental parameter set of the target user group, and obtain the optimal environmental parameter set of each user group, wherein the characteristic screening traversal solution index is used to measure the degree of adaptability between the environmental parameter indication sub-vector and the energy consumption behavior data of the user group; The energy usage behavior data prediction module is used to call the energy usage behavior data prediction model matching each of the user groups respectively, perform prediction processing according to the optimal environmental parameter set and energy usage behavior data of each of the user groups, and obtain the predicted energy usage behavior data of each of the user groups. The energy usage behavior data prediction model is composed of an encoder, a decoder, and a global attention sub-model, and the encoder and the decoder are constructed based on long-short-term neural networks.

9. A storage medium, wherein at least one executable instruction is stored in the storage medium, characterized in that: The executable instructions enable the processor to perform operations corresponding to the energy usage data prediction method based on feature screening as described in any one of claims 1-7.

10. A terminal, comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, wherein the executable instruction enables the processor to perform operations corresponding to the energy consumption data prediction method based on feature screening as described in any one of claims 1-7.

Images