A method for predicting user power consumption behavior of an electric energy meter

Abstract

The present application relates to the technical field of data processing, and particularly relates to a power meter user electricity consumption behavior prediction method, which firstly judges whether the electricity consumption behavior performance of a target user is sufficient, if the electricity consumption behavior performance is insufficient, obtains reference historical data sequences similar to the electricity consumption of the target user in a current period from historical data of other users; classifies all users according to each user feature, obtains the correlation degree of each user feature and user electricity consumption behavior according to the electricity consumption similarity of users in each category of each user feature; obtains the target category of other users corresponding to each reference historical data sequence and the target user, predicts the load data of the target user in a future period according to the correlation degree corresponding to the user feature of each target category and the data in a period after the period of each reference historical data sequence, and improves the accuracy of predicting the electricity consumption behavior of the power meter user.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a power meter user electricity consumption behavior prediction method. BACKGROUND

[0002] The power meter user electricity consumption behavior prediction refers to a process of predicting the electricity consumption data of a user in a future period of time by analyzing the historical electricity consumption data, electricity consumption habits and influencing factors of the user. According to the prediction result of the user electricity consumption behavior, the power resource scheduling can be further optimized, and the abnormal electricity consumption can be identified, which is a key of intelligent operation management of a power grid.

[0003] The traditional power meter user electricity consumption behavior prediction method mainly trains a prediction model according to the historical electricity consumption data of a current user, and predicts the electricity consumption behavior of the user according to the obtained prediction model and the current collected data. However, when the user is a new user or the electricity consumption behavior of the user changes, the prediction result is often inaccurate due to lack of enough historical data of similar electricity consumption rules, and the normal electricity consumption may be identified as abnormal electricity consumption behavior, which affects the effectiveness of power system management.

[0004] Therefore, how to improve the accuracy of predicting the electricity consumption behavior of the power meter user becomes a problem to be solved. SUMMARY

[0005] Therefore, the embodiments of the present application provide a power meter user electricity consumption behavior prediction method to solve the problem of how to improve the accuracy of predicting the electricity consumption behavior of the power meter user.

[0006] The embodiments of the present application provide a power meter user electricity consumption behavior prediction method, which comprises the following steps:

[0007] According to the data quantity and data change rule of the power load data of the target user before the current period, it is judged whether the electricity consumption behavior performance of the target user is sufficient or not;

[0008] If the electricity consumption behavior performance is not sufficient, at least one reference historical data sequence is obtained according to the similarity between the power load data of at least one other user in a preset time range before the current period and the power load data of the target user in the current period;

[0009] At least one user feature used for composing a user portrait is obtained, for any user feature, the target user and all other users are divided into at least one category according to the any user feature, and the correlation degree between the any user feature and the user electricity consumption behavior is obtained according to the change similarity of the power load data of each user in each category before the current period;

[0010] obtaining a target category in which the other users and the target user are located corresponding to each reference historical data sequence respectively, and obtaining predicted load data of the target user at each time in a future period according to a degree of association of a user feature to which each target category belongs and power load data in a period after a period in which each reference historical data sequence is located.

[0011] Preferably, the judging whether the electricity consumption behavior performance of the target user is sufficient according to the number of data and the data variation law of the power load data of the target user before the current period comprises:

[0012] If the number of data of the power load data of the target user before the current period is less than a preset number, it is determined that the electricity consumption behavior performance of the target user is insufficient.

[0013] If the number of data of the power load data of the target user before the current period is greater than or equal to the preset number, the STL algorithm is used to decompose the preset number of power load data of the target user before the current period to obtain a seasonal term and a trend term, and curve fitting is performed on the seasonal term and the trend term respectively to obtain a seasonal fitting curve and a trend fitting curve.

[0014] According to the seasonal fitting curve and the trend fitting curve, seasonal data and trend data of the target user at each time in the current period are obtained, and the seasonal data and the trend data of the target user at each time in the current period are added to obtain a load prediction value of the target user at each time in the current period.

[0015] An actual load value of the target user at each time in the current period is obtained, an absolute value of a difference between the load prediction value and the actual load value of the target user at each time in the current period is calculated to obtain a prediction error, and all prediction errors are accumulated to obtain an electricity consumption law deviation degree of the target user in the current period.

[0016] A current load sequence is composed of the actual load value of the target user at each time in the current period, an absolute value of a difference between each adjacent two data in the current load sequence is calculated, all absolute values of the differences are accumulated to obtain a data anomaly degree, an inverse of a sum of a preset constant and the data anomaly degree is calculated, and a product between the inverse and the electricity consumption law deviation degree is linearly normalized to obtain an electricity consumption behavior change degree of the target user.

[0017] If the electricity consumption behavior change degree is greater than or equal to a preset electricity consumption behavior change degree threshold, it is determined that the electricity consumption behavior performance of the target user is insufficient.

[0018] Preferably, the similarity between the power load data of the target user in the current period and the power load data of at least one other user in a preset time range before the current period is obtained to obtain at least one reference historical data sequence, including:

[0019] The power load data of each other user in a preset time range before the current period is respectively formed into a first historical sequence, and the noise degree of each data in each first historical sequence is obtained according to the data variation difference in each first historical sequence.

[0020] According to the length of the current period, the first historical sequence corresponding to each other user is respectively divided into at least one first historical sub-sequence, and the power load data of the target user at each time in the current period is formed into a current load sequence.

[0021] For any first historical sub-sequence, the similarity index between the any first historical sub-sequence and the current load sequence is obtained according to the noise degree of each data in the any first historical sub-sequence and the data similarity between the any first historical sub-sequence and the current load sequence.

[0022] The similarity indexes corresponding to all first historical sub-sequences are obtained, and the first historical sub-sequences corresponding to all similarity indexes greater than a preset similarity index threshold are recorded as reference historical data sequences.

[0023] Preferably, the noise degree of each data in each first historical sequence is obtained according to the data variation difference in each first historical sequence, including:

[0024] For any data in any first historical sequence, the neighborhood data of the two time points closest to the time point corresponding to the any data is obtained.

[0025] The absolute value of the difference between the any data and the neighborhood data of the two time points is respectively calculated to obtain a first difference degree and a second difference degree, and the sum between the first difference degree and the second difference degree is linearly normalized to obtain the noise degree of the any data.

[0026] Preferably, the similarity index between the any first historical sub-sequence and the current load sequence is obtained according to the noise degree of each data in the any first historical sub-sequence and the data similarity between the any first historical sub-sequence and the current load sequence, including:

[0027] For any data in any first historical sub-sequence, according to the position of the any data in the any first historical sub-sequence, in the current load sequence, the target data with the same position as the any data is obtained, the absolute value of the difference between the any data and the target data is calculated, the reciprocal of the sum between the absolute value of the difference and a preset constant is calculated, and the similarity between the any data and the target data is obtained;

[0028] The noise levels of all data in the any first historical sub-sequence are accumulated to obtain the overall noise level of the any first historical sub-sequence, the proportion of the noise level of the any data in the overall noise level is calculated, and the similarity confidence between the any data and the target data is obtained by subtracting the proportion from the constant 1;

[0029] The similarity confidence between the any data and the target data is taken as the weight of the similarity, and the similarity is weighted to obtain the weighted similarity between the any data and the target data;

[0030] The weighted similarities corresponding to all data in the any first historical sub-sequence are calculated, and the cumulative value of all weighted similarities is linearly normalized to obtain the similarity index between the any first historical sub-sequence and the current load sequence.

[0031] Preferably, the association degree between the any user feature and the user electricity consumption behavior is obtained according to the change similarity of the power load data of each user under each category before the current period, and the association degree between the any user feature and the user electricity consumption behavior comprises:

[0032] For any category, the power load data of each user under the any category within a preset time range before the current period is denoised to obtain a second historical sequence of each user under the any category, and each second historical sequence is divided into at least one second historical sub-sequence according to the time length of the current period;

[0033] The mean values of all data in each second historical sub-sequence are calculated, and the mean values corresponding to all second historical sub-sequences are linearly normalized to obtain the load mean value of the period in which each second historical sub-sequence is located;

[0034] The data in each second historical sub-sequence is linearly fitted to obtain a fitting straight line of each second historical sub-sequence, and the slope of the fitting straight line of each second historical sub-sequence is calculated, and all slopes are linearly normalized to obtain the load trend value of the period in which each second historical sub-sequence is located;

[0035] Based on the load mean and load trend values ​​of all second historical subsequences under any category, a scatter plot is constructed. The horizontal axis of the scatter plot represents the load mean, and the vertical axis represents the load trend value. In the scatter plot, the Euclidean distance between every two data points is calculated, and the reciprocals of all Euclidean distances are summed to obtain the accumulated value. The accumulated value is then linearly normalized to obtain the electricity consumption similarity index of all users under any category.

[0036] Calculate the average value of the similarity index of electricity consumption under all categories for any given user characteristic to obtain the degree of correlation between the given user characteristic and the user's electricity consumption behavior.

[0037] Preferably, the step of obtaining the target category of other users and the target user corresponding to each reference historical data sequence, and obtaining the predicted load data of the target user at each time in the future period based on the correlation degree of the user characteristics corresponding to each target category and the power load data in the time period after the time period of each reference historical data sequence includes:

[0038] For any reference historical data sequence, if other users and target users corresponding to any reference historical data sequence exist in any category at the same time, then the category is recorded as the target category, all target categories are obtained, and the user characteristics of all target categories are recorded as target user characteristics.

[0039] The similarity value between other users and the target user corresponding to any of the reference historical data sequences is set to a constant 1 under each target user feature, and the similarity value between other users and the target user corresponding to any of the reference historical data sequences is set to a constant 0 under each non-target user feature;

[0040] The similarity value corresponding to each user feature is used as the weight of the correlation degree corresponding to each user feature. The weighted average of the correlation degree corresponding to all user features is calculated to obtain the electricity consumption similarity between other users and the target user corresponding to any reference historical data sequence.

[0041] Based on the environmental differences between the time period corresponding to any of the reference historical data sequences and the current time period, and the electricity consumption similarity between other users and the target user corresponding to any of the reference historical data sequences, the degree to which any of the reference historical data sequences reflects the electricity consumption behavior of the target user in the current time period is obtained.

[0042] The power load data of each reference historical data sequence within a time period following that time period are combined to form a target historical sequence. Based on the data in each target historical sequence and the degree of reflection of the corresponding reference historical data sequence, the predicted load data of the target user at each time point in the future time period is obtained.

[0043] Preferably, the degree of reflection of the target user's electricity consumption behavior in the current period by any reference historical data sequence is obtained according to the difference between the corresponding period of the any reference historical data sequence and the current period, and the electricity consumption similarity between the other users corresponding to the any reference historical data sequence and the target user, comprising:

[0044] respectively obtaining the historical temperature mean value and the historical humidity mean value in the corresponding period of the any reference historical data sequence, and the real-time temperature mean value and the real-time humidity mean value in the current period;

[0045] calculating the absolute value of the difference between the historical temperature mean value and the real-time temperature mean value to obtain a temperature difference, calculating the inverse of the preset constant and the temperature difference to obtain a temperature similarity, obtaining a humidity similarity according to the historical humidity mean value and the real-time humidity mean value, and linearly normalizing the sum between the temperature similarity and the humidity similarity to obtain the environmental similarity between the corresponding period of the any reference historical data sequence and the current period;

[0046] calculating the product between the electricity consumption similarity between the other users corresponding to the any reference historical data sequence and the target user and the environmental similarity to obtain the degree of reflection of the target user's electricity consumption behavior in the current period by the any reference historical data sequence.

[0047] Preferably, the predicted load data of the target user at each time in the future period is obtained according to the data in each target historical sequence and the degree of reflection of the reference historical data sequence corresponding to each target historical sequence, comprising:

[0048] calculating the cumulative value of the degrees of reflection of the reference historical data sequences corresponding to all target historical sequences to obtain a comprehensive degree of reflection, and calculating the proportion of the degree of reflection of the reference historical data sequence corresponding to each target historical sequence in the comprehensive degree of reflection to obtain the reflection weight of each target historical sequence;

[0049] forming a time sequence for each time in the future period of the target user, for any time in the time sequence, obtaining the target historical data with the same position as the any time in each target historical sequence according to the position of the any time in the time sequence, and weighting each target historical data according to the reflection weight of the target historical sequence in which the target historical data is located to obtain the weighted average value of all target historical data, and taking the weighted average value as the predicted load data of the target user at the any time in the future period.

[0050] Preferably, the user characteristics include user electricity type, user house area, user floor, and user area.

[0051] Compared with the prior art, the embodiment of the present application has the following beneficial effects:

[0052] According to the data quantity and data variation law of the power load data of the target user before the current period, it is determined whether the electricity consumption behavior performance of the target user is sufficient; if the electricity consumption behavior performance is insufficient, at least one reference historical data sequence is obtained according to the similarity between the power load data of at least one other user in a preset time range before the current period and the power load data of the target user in the current period; at least one user characteristic used for composing a user portrait is obtained, for any user characteristic, the target user and all other users are divided into at least one category according to the any user characteristic, the correlation degree of the any user characteristic and the user electricity consumption behavior is obtained according to the variation similarity of the power load data of each user in each category before the current period; the target category in which the other user corresponding to each reference historical data sequence and the target user are located is obtained respectively, the correlation degree corresponding to the user characteristic of each target category and the power load data in a period after the period in which each reference historical data sequence is located are obtained, and the predicted load data of the target user at each time in a future period is obtained. When the electricity consumption behavior performance of the target user is insufficient, the reference historical data sequence similar to the load data variation characteristics of the target user in the current period is obtained from the historical load data of the other users, and then the load data of the target user in a future period is predicted according to the data in a period after the period in which each reference historical data sequence is located, so as to effectively solve the problem that the prediction result is inaccurate due to insufficient historical electricity consumption data of the user or change of the user electricity consumption behavior; in the process of predicting the load data of the target user in a future period according to the data in a period after the period in which each reference historical data sequence is located, the target user and all other users are classified according to the user characteristics, so as to find other users similar to the target user in user characteristics in the other users corresponding to each reference historical data sequence, and the correlation degree of each user characteristic and the electricity consumption behavior is obtained according to the similarity of the load data of the users in each category, so as to adjust the influence degree of each reference historical data sequence on the prediction result, reduce the interference of similar historical data on the prediction result caused by accidental factors, and improve the accuracy of predicting the electricity consumption behavior of the electric energy meter user. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0054] Figure 1 is a method flowchart of a power meter user electricity consumption behavior prediction method provided by the first embodiment of the present application. DETAILED DESCRIPTION

[0055] The embodiments of the present disclosure will be described in detail below, and examples of the embodiments are shown in the drawings. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present disclosure, and cannot be understood as a limitation of the present disclosure.

[0056] It should be noted that the terms "first", "second" and the like in the specification of the present disclosure and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in other than the order shown or described herein. The embodiments described in the following exemplary embodiments do not represent all the embodiments consistent with the present disclosure. Rather, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0057] In order to illustrate the technical solutions of the present application, the following will be described by specific embodiments.

[0058] Referring to Figure 1 is a method flowchart of a power meter user electricity consumption behavior prediction method provided by the first embodiment of the present application, as shown in Figure 1 , the method can include:

[0059] Step S101, according to the data quantity of the power load data of the target user before the current period and the data change rule, it is judged whether the electricity consumption behavior of the target user is sufficient.

[0060] Any user to be predicted electricity consumption behavior is recorded as a target user, and the traditional power meter user electricity consumption behavior prediction method mainly trains the prediction model according to the historical electricity consumption data of the target user, and according to the obtained prediction model, the electricity consumption behavior of the target user is predicted by using the electricity consumption data of the target user in the current period. However, when the target user is a new user, or the electricity consumption behavior of the target user changes, there is usually not enough historical data with similar electricity consumption rules, the electricity consumption behavior of the target user is insufficient, and the prediction result is inaccurate.

[0061] Therefore, firstly, it is needed to judge whether the electricity consumption behavior performance of the target user is sufficient, that is, to judge whether the target user is a new user or a user whose electricity consumption behavior has changed. If the electricity consumption behavior performance of the target user is sufficient, that is, the target user is not a new user and the electricity consumption behavior has not changed, it is indicated that the historical electricity consumption situation of the target user is sufficient to reflect the electricity consumption behavior of the target user. At this time, the electricity consumption behavior of the target user can be predicted in a traditional manner, that is, the prediction model is trained according to the historical electricity consumption data of the target user, and the electricity consumption behavior of the target user in a future time period is predicted by using the electricity consumption data of the target user in a current time period according to the obtained prediction model, wherein the time period is set to 1 hour, that is, the electricity consumption behavior of the target user in the future 1 hour is predicted according to the obtained prediction model and the data of the target user in the current 1 hour. The prediction model can be based on LSTM, XGBoost, ARIMA, etc., which is not limited here. The implementer can set it according to the specific scene. The training of the prediction model is prior art, which will not be described here.

[0062] If the electricity consumption behavior performance of the target user is not sufficient, that is, the target user is a new user or a user whose electricity consumption behavior has changed, in the embodiment of the present application, the electricity consumption behavior of the target user in a future time period is predicted according to other users with similar electricity consumption behavior as the target user under the same power supply system.

[0063] The process of judging whether the electricity consumption behavior performance of the target user is sufficient is as follows:

[0064] Firstly, it is judged whether the target user is a new user: in the embodiment of the present application, the load data of the user in the 7 days before the current time period is analyzed as historical electricity consumption data, wherein the load data is collected every minute, which is not limited here. The implementer can set the time range of the historical electricity consumption data and the data collection frequency according to the specific scene. If the electricity load data of the target user before the current time period is less than 7 days, that is, the number of the electricity load data of the target user before the current time period is less than 10080 (that is, a preset number, 7x24x60=10080), it is determined that the electricity consumption behavior performance of the target user is insufficient. It needs to be noted that considering that the general residents work, go to school and the factory operation are all in a cycle of one week, 7 days can completely reflect the electricity consumption rule of the user, and the electricity load data collected every minute corresponds to 10080 of the electricity load data collected in 7 days. Therefore, when the number of the historical electricity load data is less than 7 days for a user, it is defaulted that the user is a new user.

[0065] If the number of data of the power load data of the target user before the current period is greater than or equal to 10080, it indicates that the target user is not a new user, but it is still necessary to judge whether the power consumption behavior of the target user has changed: the power load data of the target user in the 7 days before the current period is decomposed by using the STL algorithm to obtain a seasonal term and a trend term, wherein the seasonal period length in the STL algorithm is set to 1 day, that is, 1440 data, which is not limited here, and the implementer can set it according to the specific scene. The least square method is used to respectively perform curve fitting on the seasonal term and the trend term to obtain a seasonal fitting curve and a trend fitting curve, wherein the STL algorithm and the least square method are prior art, and will not be described here.

[0066] According to the seasonal fitting curve and the trend fitting curve, the seasonal data and the trend data of each time point in the current period of the target user are obtained, and the seasonal data and the trend data of each time point in the current period of the target user are added to obtain the load prediction value of each time point in the current period of the target user. It should be noted that since the residual error decomposed by the STL algorithm is regarded as noise data, it does not participate in prediction.

[0067] The actual load value of each time point (i.e., each minute) in the current period of the target user is obtained, the absolute value of the difference between the load prediction value and the actual load value of each time point in the current period of the target user is calculated to obtain a prediction error, and all prediction errors are accumulated to obtain the power consumption rule deviation degree of the target user in the current period, which is used to judge whether the power consumption behavior of the target user has changed.

[0068] In an embodiment, the calculation formula of the power consumption rule deviation degree of the target user in the current period is:

[0069]

[0070] Wherein Q represents the power consumption rule deviation degree of the target user in the current period, N represents the number of actual load values in the current period (in the embodiment of the present application, N=60), represents the load prediction value of the i th time point in the current period, represents the actual load value of the i th time point in the current period, represents the absolute value symbol.

[0071] It should be noted that, The greater Q is, the greater the difference between the power consumption rule of the target user in the current period and the historical power consumption rule before the current period is, and the more the power consumption rule of the target user in the current period deviates from the historical power consumption rule of the target user, and thus the greater Q is, the more likely the power consumption behavior of the target user in the current period has changed.

[0072] Considering that the load data mutation caused by the electric energy meter failure will also cause a large difference between the electricity consumption rule of the target user in the current period and the historical electricity consumption rule, and the load data mutation caused by the electric energy meter failure belongs to data anomaly, not the change of the electricity consumption behavior of the user, therefore, only relying on the electricity consumption rule deviation Q cannot accurately determine whether the electricity consumption behavior of the target user has changed, and the data anomaly degree of the actual load value of the target user in the current period should also be combined to obtain the electricity consumption behavior change degree of the target user, and then the electricity consumption behavior of the target user is more accurately determined whether the electricity consumption behavior of the target user has changed, in particular:

[0073] The actual load value of the target user at each time in the current period is composed into a current load sequence, the absolute value of the difference between each adjacent two data in the current load sequence is calculated, all the absolute values of the differences are accumulated to obtain the data anomaly degree, the reciprocal of the sum between a preset constant and the data anomaly degree is calculated, and the product between the reciprocal and the electricity consumption rule deviation is linearly normalized to obtain the electricity consumption behavior change degree of the target user, wherein the linear normalization is a prior art, and will not be described here.

[0074] In an embodiment, the calculation formula of the electricity consumption behavior change degree of the target user is:

[0075]

[0076] Wherein, P represents the electricity consumption behavior change degree of the target user, Q represents the electricity consumption rule deviation of the target user in the current period, N represents the number of actual load values in the current period (in the embodiment of the present application, N=60), represents the actual load value of the i-th time in the current period, that is, the i-th actual load value in the current load sequence, represents the actual load value of the i-1-th time in the current period, that is, the i-1-th actual load value in the current load sequence, represents a preset constant, which is used to prevent the denominator from being 0, in the embodiment of the present application, is set to This is not limited here, and the implementer can set it according to the specific scene, represents the absolute value symbol, represents the linear normalization function.

[0077] It should be noted that, The smaller Q is, the larger Q is, which means that the possibility of abnormal data caused by the electric energy meter failure in the current load sequence is smaller, the difference between the electricity consumption rule of the target user in the current period and the historical electricity consumption rule before the current period is larger, the authenticity of the change of the electricity consumption behavior of the target user in the current period is larger, and then P is larger, and the electricity consumption behavior of the target user in the current period is more likely to have changed.

[0078] Whether the user's electricity consumption behavior changes is greatly different in the electricity consumption data, and the two cases of the electricity consumption behavior change degree after normalization are divided on both sides of the interval [0, 1], so the intermediate value is taken as the division threshold, that is, the preset electricity consumption behavior change degree threshold is set to 0.5, which is not limited here, and the implementer can set it according to the specific scene. If P≥0.5, it is determined that the electricity consumption behavior of the target user in the current period has changed, and the electricity consumption behavior of the target user is not sufficient.

[0079] At this point, the judgment of whether the electricity consumption behavior of the target user is sufficient is completed.

[0080] In step S102, if the electricity consumption behavior is not sufficient, at least one reference historical data sequence is obtained according to the similarity between the power load data of at least one other user in a preset time range before the current period and the power load data of the target user in the current period.

[0081] If the electricity consumption behavior of the target user is not sufficient, that is, the target user is a new user or a user whose electricity consumption behavior has changed, in the embodiment of the present application, at least one reference historical data sequence similar to the electricity consumption behavior of the target user in the current period is obtained from the power load data of each other user in the target user in the power supply system within 7 days before the current period (not limited here, the implementer can set the time range according to the specific scene), which is used for subsequent prediction of the electricity consumption behavior of the target user in the future period according to the load data in the next period of each reference historical data sequence.

[0082] Since the power load data of each other user within 7 days before the current period may contain noise data, these noise data will affect the judgment of similarity, so first, the power load data of each other user within 7 days before the current period is grouped into a first historical sequence, and the noise degree of each data in each first historical sequence is obtained according to the data change difference in each first historical sequence, so as to reduce the noise influence in the subsequent data analysis process. Taking the jth data in the first historical sequence of the uth other user as an example, the specific way to obtain the noise degree of the jth data in the first historical sequence of the uth other user is as follows:

[0083] In the first historical sequence of the uth other user, the neighborhood data of the two time points closest to the jth data is obtained.

[0084] respectively, to obtain a first difference degree and a second difference degree, linearly normalizing a sum between the first difference degree and the second difference degree to obtain the noise degree of the jth data in the first historical sequence of the u th other user.

[0085] In an embodiment, a calculation formula of the noise degree of the jth data in the first historical sequence of the u th other user is:

[0086]

[0087] wherein, denotes the noise degree of the jth data in the first historical sequence of the u th other user, denotes a first neighborhood data in the two neighborhood data closest to a time corresponding to the jth data, denotes a second neighborhood data in the two neighborhood data closest to a time corresponding to the jth data, denotes the jth data in the first historical sequence of the u th other user, denotes an absolute value symbol, denotes a linear normalization function.

[0088] It should be noted that, the greater the jth data is, the greater the mutation degree of the jth data is, and further the greater the jth data is, the more likely the jth data is noise data.

[0089] Similarly, the noise degree of each data in the first historical sequence of each other user is obtained.

[0090] Further, according to a length of a current period, i.e. a length of 1 hour, each first historical sequence corresponding to each other user is divided into at least one first historical sub-sequence (in an embodiment of the present application, each first historical sequence corresponding to each other user is power load data in the past 7 days before the current 1 hour, so the number of each first historical sub-sequence corresponding to each other user is 168), according to a current load sequence of the target user in the current period, the similarity between each first historical sub-sequence and the current load sequence is analyzed to obtain a similarity index between each first historical sub-sequence and the current load sequence, and further at least one reference historical data sequence similar to the power consumption behavior of the target user in the current period is found.

[0091] Taking the vth first historical subsequence of the u th other user as an example, according to the noise degree of each data in the vth first historical subsequence of the u th other user and the data similarity between the vth first historical subsequence of the u th other user and the current load sequence, a similarity index between the vth first historical subsequence of the u th other user and the current load sequence is obtained, and specifically:

[0092] For any data in the vth first historical subsequence of the u th other user, according to the position of the any data in the vth first historical subsequence of the u th other user, a target data with the same position as the any data is obtained in the current load sequence, the absolute value of the difference between the any data and the target data is calculated, the reciprocal of the sum of the absolute value of the difference and a preset constant is calculated, and the similarity degree between the any data and the target data is obtained.

[0093] The noise degrees of all data in the vth first historical subsequence of the u th other user are accumulated to obtain the overall noise degree of the vth first historical subsequence of the u th other user, the proportion of the noise degree of the any data in the overall noise degree is calculated, and the similarity confidence between the any data and the target data is obtained by subtracting the proportion from the constant 1.

[0094] The similarity confidence between the any data and the target data is taken as the weight of the similarity degree, and the similarity degree is weighted to obtain the weighted similarity between the any data and the target data.

[0095] The weighted similarities corresponding to all data in the vth first historical subsequence of the u th other user are calculated, and the cumulative value of all weighted similarities is linearly normalized to obtain the similarity index between the vth first historical subsequence of the u th other user and the current load sequence.

[0096] In an embodiment, the calculation formula of the similarity index between the vth first historical subsequence of the u th other user and the current load sequence is:

[0097]

[0098] wherein, represents the similarity index between the vth first historical subsequence of the u th other user and the current load sequence, N represents the number of data in the current load sequence (in the embodiment of the present application, N = 60), the number of data in the vth first historical subsequence of the u th other user is equal to the number of data in the current load sequence, represents the z th data in the vth first historical subsequence of the u th other user, denotes the zth data in the current load sequence, denotes a preset constant for preventing the denominator from being 0, in the embodiment of the present application, is set to 1 Here, no limitation is made, and the implementer can set it according to the specific scene, denotes the noise degree of the jth data in the vth first historical sub-sequence of the uth other user, denotes the noise degree of the zth data in the vth first historical sub-sequence of the uth other user, denotes the absolute value symbol, denotes a linear normalization function.

[0099] It should be noted that, The smaller the value is, the greater the similarity between the vth first historical sub-sequence of the uth other user and the current load sequence is, The smaller the value is, the more credible the similarity between the vth first historical sub-sequence of the uth other user and the current load sequence is, and further The greater the value is, the more similar the vth first historical sub-sequence of the uth other user and the current load sequence are, and the more similar the power consumption behavior of the uth other user in the time period in which the vth first historical sub-sequence is located and the power consumption behavior of the target user in the current time period are.

[0100] Similarly, the similarity index between each first historical sub-sequence of each other user and the current load sequence is obtained, according to the abnormal data distribution in the experimental data, it is found that the abnormal data and the normal data are usually distributed on both sides of 0.7, therefore, the preset similarity index threshold is set to 0.7, here, no limitation is made, and the implementer can set it according to the specific scene, the first historical sub-sequence corresponding to all similarity indexes greater than 0.7 is obtained, and is recorded as a reference historical data sequence similar to the power consumption behavior of the target user in the current time period, which is used for subsequent prediction of the power consumption behavior of the target user in the future time period according to the load data in the time period after the time period in which each reference historical data sequence is located.

[0101] In step S103, at least one user feature used to constitute a user portrait is obtained, for any user feature, according to the any user feature, the target user and all other users are divided into at least one category, and the correlation degree of the any user feature and the user power consumption behavior is obtained according to the change similarity of the power load data of each user in each category before the current time period.

[0102] The reference historical data sequence similar to the power consumption behavior of the target user in the current period is obtained through step S102, but since the reference historical data sequence represents the power consumption of other users in a historical period, even if the power consumption of other users in the historical period is similar to the power consumption of the target user in the current period, the power consumption in the future is likely to be different. Considering that the power consumption behaviors of users with different user characteristics are likely to be different, for example, for apartment housing, the 1st to 3rd floors are generally commercial power, and the 4th floor and above are living power. The living power has a power consumption peak in the morning and evening, while the commercial power may be stable throughout the day; another example is an industrial production workshop. The large industrial production workshop generally has more production equipment than the small industrial production workshop, and the large industrial production workshop also has more power consumption than the small industrial production workshop.

[0103] Therefore, in the embodiment of the present application, first, the user power consumption type (such as living power, general commercial power, agricultural production power, etc.), the user house area (such as small, medium, large, etc.), the user floor (such as low, medium, high, etc.), and the user area are used to form the user characteristics of the user portrait, and all users under the power supply system of the target user are divided into at least one category according to each user characteristic. If any two users belong to the same category, it means that the two users have similar user characteristics. Among them, all users under the power supply system of the target user are divided into at least one category according to each user characteristic. For example, according to the user power consumption type, all users are divided into living power users, general commercial power users, agricultural production power users, etc.; according to the user house area, all users are divided into small house users (for example, less than 60 square meters for small house), medium house users (for example, 60 square meters to 120 square meters for medium house), large house users (for example, 120 square meters to 200 square meters for large house), etc.; according to the user floor, all users are divided into low floor users (for example, 1st to 3rd floor for low floor users), medium floor users (for example, 4th to 12th floor for medium floor users), high floor users (for example, 13th floor and above for high floor users), etc.; according to the user area, the latitude and longitude of the location of all users are mapped into a scatter plot. The horizontal coordinate of the scatter plot represents the longitude, and the vertical coordinate represents the latitude. The data points in the scatter plot are clustered using the DBSCAN clustering algorithm to obtain at least one clustering cluster. The users belonging to the same clustering cluster are considered as users in the same area. Users with houses facing the same direction can also be considered as users in the same area, or all users can be classified according to the official division of the area. Here, the number of user characteristics and the classification method and classification standard of each user characteristic can be set according to the specific scene by the implementer. The DBSCAN clustering algorithm is a prior art, which will not be described here.

[0104] Then, according to the change similarity of the power load data of the users belonging to the same category before the current period under different user characteristics, the association degree of each user characteristic and the user power consumption behavior is obtained. If the association degree of a certain user characteristic is large, it indicates that the power consumption behaviors of similar users under the same user characteristic are similar. That is, if any reference historical data sequence corresponds to other users and the target user belong to the same category under any user characteristic, and the association degree of any user characteristic is large, it indicates that the other users corresponding to any reference historical data sequence and the target user have similar power consumption behaviors. It can be considered that the power consumption behavior in the period of any reference historical data sequence is not only similar to the power consumption behavior of the target user in the current period, but also very likely similar to the power consumption behavior of the target user in the future period. Therefore, when predicting the power consumption behavior of the target user in the future period according to the data of the period after the period of any reference historical data sequence, a higher weight should be given to it to adjust the influence degree of any reference historical data sequence on the prediction result, thereby improving the accuracy of predicting the power consumption behavior of the target user in the future period.

[0105] For the association degree of each user characteristic and the user power consumption behavior, taking the a-th user characteristic as an example, the association degree of the a-th user characteristic and the user power consumption behavior is obtained according to the change similarity of the power load data of all users under each category of the a-th user characteristic before the current period, and the association degree of the a-th user characteristic and the user power consumption behavior is obtained. Specifically:

[0106] (1) According to the change similarity of the power load data of the users under each category of the a-th user characteristic before the current period, the power consumption similarity index of all users under each category of the a-th user characteristic is obtained.

[0107] Specifically: for the t-th category of the a-th user characteristic, the power load data of each user under the t-th category within 7 days before the current period (here, the time range is not limited, and the implementer can set the time range according to the specific scene) is denoised by using median filtering to obtain the second historical sequence of each user under the t-th category. According to the time length of the current period (in the embodiment of the present application, the time length of the current period is 1 hour), each second historical sequence is divided into at least one second historical sub-sequence. The denoising processing by using median filtering is a prior art, which will not be described here.

[0108] The mean value of all data in each second historical sub-sequence is calculated, and the mean values corresponding to all second historical sub-sequences are linearly normalized to obtain the load mean value of the period of each second historical sub-sequence.

[0109] The data in each of the second historical sub-sequences is fitted with a straight line by least square method to obtain a fitting straight line of each of the second historical sub-sequences, the slope of the fitting straight line of each of the second historical sub-sequences is calculated respectively, and all the slopes are linearly normalized to obtain a load trend value of the time period in which each of the second historical sub-sequences is located;

[0110] According to the load mean and the load trend value of all the second historical sub-sequences under the t-th category, a scatter plot is constructed, the horizontal coordinate of the scatter plot represents the load mean, and the vertical coordinate represents the load trend value, in the scatter plot, the Euclidean distance between each two data points is calculated, that is, the data in the scatter plot is grouped two by two, the Euclidean distance between the two data in each group is calculated, the reciprocal of all the Euclidean distances is accumulated to obtain an accumulated value, and the accumulated value is linearly normalized to obtain the electricity consumption situation similarity index of all the users under the t-th category.

[0111] In an embodiment, the calculation formula of the electricity consumption situation similarity index of all the users under the t-th category is:

[0112]

[0113] wherein, represents the electricity consumption situation similarity index of all the users under the t-th category, represents the number of all the data points in the scatter plot constructed according to the load mean and the load trend value of all the second historical sub-sequences under the t-th category, that is, the number of all the second historical sub-sequences under the t-th category, that is, the number of combinations of all the data in the scatter plot two by two, represents the Euclidean distance between the two data in the c-th combination, represents a linear normalization function.

[0114] It should be noted that, The smaller the accumulated value is, the more densely the data in the scatter plot is distributed, the more similar the load data features of the users under the t-th category are, and further The larger the accumulated value is, the more similar the electricity consumption situations of the users under the t-th category are.

[0115] Similarly, the electricity consumption situation similarity index under each category of the a-th user feature is obtained.

[0116] (2) According to the electricity consumption situation similarity index under each category of the a-th user feature, the correlation degree of the a-th user feature and the user electricity consumption behavior is obtained.

[0117] Specifically, the average value of the electricity consumption situation similarity indexes of all the categories under the a-th user feature is calculated to obtain the correlation degree of the a-th user feature and the user electricity consumption behavior.

[0118] In an embodiment, the formula for calculating the correlation degree of the a-th user feature and the user's power consumption behavior is:

[0119]

[0120] wherein, represents the correlation degree of the a-th user feature and the user's power consumption behavior, represents the number of all categories under the a-th user feature, represents the power consumption similarity index under the t-th category of the a-th user feature.

[0121] It should be noted that, the greater the value is, the more similar the power consumption of the user under each category of the a-th user feature is, and further the greater the value is, the greater the correlation degree of the a-th user feature and the user's power consumption behavior is.

[0122] Similarly, the correlation degree of each user feature and the user's power consumption behavior is obtained.

[0123] In step S104, the target category in which the target user and the other users corresponding to each reference historical data sequence are located is obtained respectively, the predicted load data of the target user at each time in the future period is obtained according to the correlation degree corresponding to the user feature to which each target category belongs and the power load data in a period after the period in which each reference historical data sequence is located.

[0124] The association degree of each user feature and user electricity behavior is obtained through step S103, and the reference historical data sequence obtained in step S102 is historical data similar to the electricity consumption of the target user in the current period within a certain historical period. If any reference historical data sequence corresponds to other users and the target user belonging to the same category under a certain user feature, and the association degree of the user feature and the user electricity behavior is large, it is considered that the electricity behavior in a period after the period corresponding to any reference historical data sequence is also likely to be similar to the electricity behavior of the target user in the future period. The greater the degree of reflection of any reference historical data sequence on the electricity behavior of the target user, the greater the weight given to it when predicting the electricity behavior of the target user in the future period based on the data of the period after the period of any reference historical data sequence. At the same time, considering that changes in temperature and humidity and other environments are also one of the important factors affecting user electricity consumption, the closer the temperature and humidity of the period of any reference historical data sequence to the current period, the greater the degree of reflection of any reference historical data sequence on the electricity behavior of the target user, and then when predicting the electricity behavior of the target user in the future period based on the data of the period after the period of any reference historical data sequence, it should also be given a higher weight to improve the accuracy of predicting the electricity behavior of the target user in the future period.

[0125] For the degree of reflection of the reference historical data sequence on the electricity behavior of the target user: taking the vth reference historical data sequence as an example, if any category simultaneously exists the other user corresponding to the vth reference historical data sequence and the target user, then any category is recorded as a target category, all target categories are obtained, and the user features belonging to all target categories are recorded as target user features, that is, if the other user corresponding to the vth reference historical data sequence and the target user are classified into a category under a certain user feature, the user feature is recorded as a target user feature. According to the association degree corresponding to each target user feature and the environmental difference between the period corresponding to the vth reference historical data sequence and the current period, the degree of reflection of the vth reference historical data sequence on the electricity behavior of the target user in the current period is obtained, and the specific steps are as follows:

[0126] (1) According to the association degree corresponding to each target user feature, the electricity similarity between the other user corresponding to the vth reference historical data sequence and the target user is obtained.

[0127] Specifically: the similarity value between the other user corresponding to the vth reference historical data sequence and the target user under each target user feature is set to a constant 1, and the similarity value between the other user corresponding to the vth reference historical data sequence and the target user under each non-target user feature is set to a constant 0.

[0128] The similarity value corresponding to each user feature is taken as the weight of the corresponding correlation degree of each user feature, a weighted mean of the corresponding correlation degrees of all user features is calculated, and the power consumption similarity between the target user and other users corresponding to the vth reference historical data sequence is obtained.

[0129] In an embodiment, the calculation formula of the power consumption similarity between the target user and other users corresponding to the vth reference historical data sequence is:

[0130]

[0131] wherein, represents the power consumption similarity between the target user and other users corresponding to the vth reference historical data sequence, represents the number of user features, in the embodiment of the present application, , respectively, are the user power consumption type, the user house area, the user floor, and the user area, represents the similarity value (1 or 0) between the target user and other users corresponding to the vth reference historical data sequence under the a th user feature, represents the correlation degree of the a th user feature and the user power consumption behavior.

[0132] It should be noted that, The greater the value is, the more similar the power consumption behaviors of the target user and other users corresponding to the vth reference historical data sequence are, and thus The greater the weight corresponding to the vth reference historical data sequence is.

[0133] (2) According to the power consumption similarity between the target user and other users corresponding to the vth reference historical data sequence, and the environmental difference between the period corresponding to the vth reference historical data sequence and the current period, the reflection degree of the power consumption behavior of the target user in the current period is obtained.

[0134] Specifically: through the meteorological platform of the region where the power supply system belongs, the temperature value and the humidity value of each time in the period corresponding to the vth reference historical data sequence and the current period are obtained according to the acquisition frequency of the load data in step S101, i.e. once per minute, and all the obtained temperature values and humidity values are linearly normalized to obtain the corresponding temperature data and humidity data. According to the temperature data and humidity data of each time in the period corresponding to the vth reference historical data sequence and the current period, the historical temperature mean value and the historical humidity mean value in the period corresponding to the vth reference historical data sequence, and the real-time temperature mean value and the real-time humidity mean value in the current period are calculated respectively.

[0135] calculating an absolute value of a difference between the historical temperature mean value and the real-time temperature mean value to obtain a temperature difference value, calculating an inverse of a preset constant and the temperature difference value to obtain a temperature similarity, obtaining a humidity similarity according to the historical humidity mean value and the real-time humidity mean value, performing linear normalization on a sum between the temperature similarity and the humidity similarity to obtain an environment similarity between a time period corresponding to the vth reference historical data sequence and a current time period;

[0136] calculating a product between the electricity consumption similarity between the other users corresponding to the vth reference historical data sequence and the target user and the environment similarity to obtain a reflection degree of the vth reference historical data sequence on the electricity consumption behavior of the target user in the current time period.

[0137] In an embodiment, a calculation formula of the reflection degree of the vth reference historical data sequence on the electricity consumption behavior of the target user in the current time period is as follows:

[0138]

[0139] wherein, represents the reflection degree of the vth reference historical data sequence on the electricity consumption behavior of the target user in the current time period, represents the electricity consumption similarity between the other users corresponding to the vth reference historical data sequence and the target user, represents the historical temperature mean value in the time period corresponding to the vth reference historical data sequence, represents the real-time temperature mean value in the current time period, represents the historical humidity mean value in the time period corresponding to the vth reference historical data sequence, represents the real-time humidity mean value in the current time period, represents a preset constant, which is used to prevent the denominator from being 0, and in an embodiment of the present application, is set to 1, which is not limited here, and can be set according to a specific scene by an implementer, represents an absolute value symbol, represents a linear normalization function.

[0140] It should be noted that, the smaller the environment similarity is, the smaller the electricity consumption similarity is, the greater the environment similarity is, the more similar the electricity consumption behaviors of the other users corresponding to the vth reference historical data sequence and the target user are, and thus the greater the reflection degree of the vth reference historical data sequence on the electricity consumption behavior of the target user in the current time period is, the greater the weight of the vth reference historical data sequence is.

[0141] Thus, we have obtained the degree to which the v-th reference historical data sequence reflects the target user's electricity consumption behavior in the current time period. Similarly, we obtain the degree to which each reference historical data sequence reflects the target user's electricity consumption behavior in the current time period.

[0142] Furthermore, based on the data from the time period following each reference historical data sequence, the electricity consumption behavior of the target user in the future time period is predicted. First, the electricity load data for the time period following each reference historical data sequence (1 hour) is used to form target historical sequences. Then, based on the data in each target historical sequence and the reflectivity of the corresponding reference historical data sequence, the load data of the target user at each moment in the future time period (i.e., the next 1 hour) is predicted, resulting in the predicted load data for the target user at each moment in the future time period. Specifically:

[0143] Calculate the cumulative value of the reflection degree of the reference historical data sequences corresponding to all target historical sequences to obtain the comprehensive reflection degree. Calculate the proportion of the reflection degree of the reference historical data sequences corresponding to each target historical sequence in the comprehensive reflection degree to obtain the reflection weight of each target historical sequence.

[0144] The target user is organized into a time series for each moment in the future period. For any moment in the time series, target historical data with the same position as the target moment is obtained in each target historical sequence according to the position of the target historical sequence in which the target historical data is located. Each target historical data is weighted according to the reflection weight of the target historical sequence in which the target historical data is located, and the weighted average of all target historical data is obtained. The weighted average is used as the predicted load data of the target user at any moment in the future period.

[0145] In one embodiment, taking the x-th moment in the time series as an example, that is, taking the x-th moment in a future time period as an example, the formula for calculating the predicted load data of the target user at the x-th moment in the future time period is as follows:

[0146]

[0147] in, This represents the predicted load data of the target user at time x within a future time period, where m represents the number of reference historical data sequences, which is also the number of target historical sequences. This represents the x-th data point in the target historical sequence corresponding to the v-th reference historical data sequence. This indicates the degree of reflection of the v-th reference historical data sequence. Indicates the first The degree of reflection of a reference historical data sequence.

[0148] Similarly, the predicted load data of the target user at each time in the future period is obtained, and the prediction of the power consumption behavior of the target user in the future period is completed.

[0149] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for predicting electricity consumption behavior of electricity meter users, characterized in that, The method for predicting electricity consumption behavior of electricity meter users includes: Based on the amount of electricity load data of the target user before the current time period and the data change pattern, determine whether the target user's electricity consumption behavior is sufficient; If the electricity consumption behavior is insufficient, at least one reference historical data sequence is obtained based on the similarity between the electricity load data of at least one other user within a preset time range before the current time period and the electricity load data of the target user within the current time period. Obtain at least one user feature to form a user profile. For any user feature, classify the target user and all other users into at least one category based on the user feature. Based on the similarity of the changes in power load data of each user in each category before the current time period, obtain the degree of correlation between the user feature and the user's electricity consumption behavior. The target category of other users and the target user corresponding to each reference historical data sequence is obtained respectively. Based on the correlation degree of the user characteristics of each target category and the power load data of the time period after the time period of each reference historical data sequence, the predicted load data of the target user at each time in the future time period is obtained. The step of determining whether the target user's electricity consumption behavior is sufficient based on the amount of electricity load data and the data change patterns of the target user before the current time period includes: If the number of data points for the target user's electricity load prior to the current time period is less than the preset number, then the target user's electricity consumption behavior is determined to be insufficient. If the number of power load data for the target user before the current time period is greater than or equal to a preset number, the STL algorithm is used to decompose the preset number of power load data for the target user before the current time period to obtain seasonal items and trend items. Curve fitting is then performed on the seasonal items and trend items to obtain seasonal fitting curves and trend fitting curves. Based on the seasonal fitting curve and the trend fitting curve, the seasonal data and trend data of the target user at each moment in the current period are obtained. The seasonal data and trend data of the target user at each moment in the current period are added together to obtain the load forecast value of the target user at each moment in the current period. Obtain the actual load value of the target user at each moment in the current time period, calculate the absolute value of the difference between the predicted load value and the actual load value of the target user at each moment in the current time period, obtain the prediction error, and accumulate all prediction errors to obtain the deviation of the target user's electricity consumption pattern in the current time period. The actual load values ​​of the target user at each moment in the current period are used to form the current load sequence. The absolute value of the difference between each two adjacent data in the current load sequence is calculated. All absolute values ​​of the difference are accumulated to obtain the degree of data anomaly. The reciprocal of the sum between the preset constant and the degree of data anomaly is calculated. The product of the reciprocal and the deviation of the electricity consumption pattern is linearly normalized to obtain the degree of change in the electricity consumption behavior of the target user. If the degree of change in electricity consumption behavior is greater than or equal to a preset threshold for the degree of change in electricity consumption behavior, then the target user's electricity consumption behavior is determined to be insufficient. The step of obtaining the target category of other users and the target user corresponding to each reference historical data sequence, and obtaining the predicted load data of the target user at each time in the future period based on the correlation degree of user characteristics to which each target category belongs and the power load data in the time period following each reference historical data sequence, includes: For any reference historical data sequence, if other users and target users corresponding to any reference historical data sequence exist in any category at the same time, then the category is recorded as the target category, all target categories are obtained, and the user characteristics of all target categories are recorded as target user characteristics. The similarity value between other users and the target user corresponding to any of the reference historical data sequences is set to a constant 1 under each target user feature, and the similarity value between other users and the target user corresponding to any of the reference historical data sequences is set to a constant 0 under each non-target user feature; The similarity value corresponding to each user feature is used as the weight of the correlation degree corresponding to each user feature. The weighted average of the correlation degree corresponding to all user features is calculated to obtain the electricity consumption similarity between other users and the target user corresponding to any reference historical data sequence. Based on the environmental differences between the time period corresponding to any of the reference historical data sequences and the current time period, and the electricity consumption similarity between other users and the target user corresponding to any of the reference historical data sequences, the degree to which any of the reference historical data sequences reflects the electricity consumption behavior of the target user in the current time period is obtained. The power load data of each reference historical data sequence within a time period following that time period are combined to form a target historical sequence. Based on the data in each target historical sequence and the degree of reflection of the corresponding reference historical data sequence, the predicted load data of the target user at each time point in the future time period is obtained.

2. The method for predicting user electricity consumption behavior according to claim 1, characterized in that, The step of obtaining at least one reference historical data sequence based on the similarity between the power load data of at least one other user within a preset time range prior to the current time period and the power load data of the target user within the current time period includes: The power load data of each other user within a preset time range before the current time period are respectively composed into a first historical sequence. Based on the differences in data changes in each first historical sequence, the noise level of each data in each first historical sequence is obtained. Based on the length of the current time period, the first historical sequence corresponding to each other user is divided into at least one first historical subsequence, and the power load data of the target user at each moment in the current time period is used to form the current load sequence. For any first historical subsequence, based on the noise level of each data point in the first historical subsequence and the data similarity between the first historical subsequence and the current load sequence, a similarity index between the first historical subsequence and the current load sequence is obtained; Obtain the similarity index corresponding to all first historical subsequences, and record the first historical subsequences corresponding to all similarity indices that are greater than the preset similarity index threshold as reference historical data sequences.

3. The method for predicting electricity consumption behavior of electricity meter users according to claim 2, characterized in that, The step of obtaining the noise level of each data point in each of the first historical sequences based on the differences in data changes in each of the first historical sequences includes: For any data in any first historical sequence, obtain the neighborhood data of the two closest times to the time corresponding to the data. Calculate the absolute value of the difference between any data and the neighborhood data at the two time points to obtain the first degree of difference and the second degree of difference. Linearly normalize the sum of the first degree of difference and the second degree of difference to obtain the noise level of any data.

4. The method for predicting user electricity consumption behavior according to claim 2, characterized in that, The step of obtaining a similarity index between the first historical subsequence and the current load sequence based on the noise level of each data point in the first historical subsequence and the data similarity between the first historical subsequence and the current load sequence includes: For any data in any first historical subsequence, based on the position of any data in any first historical subsequence, in the current load sequence, obtain target data with the same position as any data, calculate the absolute value of the difference between any data and the target data, calculate the reciprocal of the sum of the absolute value of the difference and a preset constant, and obtain the similarity between any data and the target data. The noise levels of all data in any first historical subsequence are summed to obtain the overall noise level of any first historical subsequence. The proportion of the noise level of any data in the overall noise level is calculated. The constant 1 is subtracted from the proportion to obtain the similarity confidence between any data and the target data. The similarity confidence between any data and the target data is used as the weight of the similarity degree, and the similarity degree is weighted to obtain the weighted similarity between any data and the target data; Calculate the weighted similarity of all data in any first historical subsequence, and linearly normalize the sum of all weighted similarities to obtain the similarity index between any first historical subsequence and the current load sequence.

5. The method for predicting user electricity consumption behavior according to claim 1, characterized in that, The method of determining the correlation between any user characteristic and user electricity consumption behavior based on the similarity of changes in electricity load data for each user in each category before the current time period includes: For any category, the power load data of each user in any category within a preset time range before the current time period is denoised to obtain the second historical sequence of each user in any category. According to the time length of the current time period, each second historical sequence is divided into at least one second historical subsequence. Calculate the mean of all data in each second historical subsequence, and linearly normalize the mean of all second historical subsequences to obtain the load mean of each second historical subsequence in the time period. Linear fitting is performed on the data in each second historical subsequence to obtain the fitted line of each second historical subsequence. The slope of the fitted line of each second historical subsequence is calculated. All slopes are linearly normalized to obtain the load trend value of each second historical subsequence in the time period. Based on the load mean and load trend values ​​of all second historical subsequences under any category, a scatter plot is constructed. The horizontal axis of the scatter plot represents the load mean, and the vertical axis represents the load trend value. In the scatter plot, the Euclidean distance between every two data points is calculated, and the reciprocals of all Euclidean distances are summed to obtain the accumulated value. The accumulated value is then linearly normalized to obtain the electricity consumption similarity index of all users under any category. Calculate the average value of the similarity index of electricity consumption under all categories for any given user characteristic to obtain the degree of correlation between the given user characteristic and the user's electricity consumption behavior.

6. The method for predicting user electricity consumption behavior according to claim 1, characterized in that, The step of obtaining the degree to which any reference historical data sequence reflects the target user's electricity consumption behavior in the current period based on the environmental differences between the corresponding time period and the current time period, and the similarity of electricity consumption between other users and the target user corresponding to any reference historical data sequence, includes: Obtain the historical average temperature and historical average humidity within the corresponding time period of any of the reference historical data sequences, as well as the real-time average temperature and real-time average humidity within the current time period. Calculate the absolute value of the difference between the historical average temperature and the real-time average temperature to obtain the temperature difference value. Calculate the reciprocal between a preset constant and the temperature difference value to obtain the temperature similarity. Obtain the humidity similarity value based on the historical average humidity and the real-time average humidity. Linearly normalize the sum of the temperature similarity value and the humidity similarity value to obtain the environmental similarity between the time period corresponding to any reference historical data sequence and the current time period. Calculate the product between the electricity consumption similarity of other users and the target user corresponding to any reference historical data sequence and the environmental similarity to obtain the degree to which the reference historical data sequence reflects the electricity consumption behavior of the target user in the current time period.

7. The method for predicting electricity consumption behavior of electricity meter users according to claim 1, characterized in that, The step of obtaining the predicted load data of the target user at each time point in the future period based on the data in each target historical sequence and the degree of reflection of the corresponding reference historical data sequence includes: Calculate the cumulative value of the reflection degree of the reference historical data sequences corresponding to all target historical sequences to obtain the comprehensive reflection degree. Calculate the proportion of the reflection degree of the reference historical data sequences corresponding to each target historical sequence in the comprehensive reflection degree to obtain the reflection weight of each target historical sequence. The target user is organized into a time series for each moment in the future period. For any moment in the time series, target historical data with the same position as the target moment is obtained in each target historical sequence according to the position of the target historical sequence in which the target historical data is located. Each target historical data is weighted according to the reflection weight of the target historical sequence in which the target historical data is located, and the weighted average of all target historical data is obtained. The weighted average is used as the predicted load data of the target user at any moment in the future period.

8. The method for predicting user electricity consumption behavior according to claim 1, characterized in that, The user characteristics include the user's electricity consumption type, the size of the user's house, the floor the user is on, and the area the user is in.

Images