User electricity consumption type classification system and method

Abstract

The invention provides a user electricity consumption pattern classification system and method, and the method comprises the steps: carrying out the first classification through a first machine learning grouping algorithm according to the electricity consumption feature information of a training electricity consumption data group after reading the training electricity consumption data group, and generating a plurality of first-order electricity consumption data groups; then, the power consumption data group is further subjected to a feature extraction algorithm to generate a second-order feature value, and the second-order feature value is classified for the second time by a second machine learning grouping algorithm to generate a second-order power consumption data group in the first-order power consumption data group. When a to-be-classified power utilization data set is received, a power utilization type classification result is determined according to the average similarity between the to-be-classified power utilization data set and each second-order power utilization data set. The second-order power consumption data group generated by the second-stage classification can accurately represent different power consumption forms of the power consumption data group, so that an accurate power consumption form classification result is provided.

Description

Technical Field

[0001] A classification system and method, in particular, a classification system and method for user electricity usage patterns. Background Art

[0002] Energy demand is increasing day by day. While developing energy technology to increase energy supply, how to save energy and reduce consumption is also an important part of improving energy efficiency. In order to reduce unnecessary energy consumption and allocate electricity efficiently, the latest energy-saving trend in the world in recent years is to use cloud platforms, behavioral science and power big data analysis technology to analyze the electricity behavior of household users or other individual electricity users, and provide individual improvement suggestions based on the analysis results to encourage household users or individual electricity users to improve their electricity usage methods to achieve energy-saving goals.

[0003] In the existing power consumption analysis technology, generally the power consumption load ratio of multiple time units (e.g., half an hour) of an individual power user in a time period (e.g., a day) is plotted into a power consumption load curve, and the power consumption behavior of different time periods is analyzed according to the shape and composition of the curve to generate analysis results. For example, the power consumption ratio of this individual power user is low in the morning and afternoon, and high in the evening, and improvement or adjustment suggestions are provided accordingly.

[0004] However, the existing power consumption analysis technology has few analysis factors, and the analysis results can only roughly compare the power consumption ratios in different time periods, and the power consumption adjustment suggestions it can provide are limited. In addition, how to integrate the analysis results of a large number of individual power users and use big data for integrated analysis to provide energy industry improvement solutions is an issue that needs to be further resolved in the development of energy technology. Summary of the invention

[0005] In view of the shortcomings of the existing power consumption analysis technology, the present invention provides a user power consumption type classification system, including:

[0006] A storage device stores a plurality of training power consumption data sets, each of which includes a plurality of unit time power consumption values ​​having time sequence dependencies, and each of which has power consumption characteristic information;

[0007] a processing device connected to the storage device, and reading the plurality of training electrical data sets from the storage device;

[0008] The processing device uses the power consumption characteristic information of the plurality of training power consumption data groups as the first-order characteristic value, and classifies the plurality of training power consumption data groups into a plurality of first-order power consumption data groups through a first machine learning clustering algorithm according to a first clustering quantity value;

[0009] The processing device calculates at least one second-order feature value of the training power consumption data group in the plurality of first-order power consumption data groups respectively through a feature extraction algorithm;

[0010] The processing device classifies the plurality of training power consumption data groups into a plurality of second-order power consumption data groups according to the at least one second-order feature value of each training power consumption data group and a second clustering quantity value through a second machine learning clustering algorithm, and assigns a power consumption type label to each second-order power consumption data group;

[0011] The processing device receives a power consumption data group to be classified, calculates an average similarity between the power consumption data group to be classified and the training power consumption data groups in each of the second-order power consumption data groups, selects the power consumption type label of the second-order power consumption data group with the highest average similarity as a power consumption type classification result of the power consumption data group to be classified, and outputs the power consumption type classification result.

[0012] In addition, the present invention also provides a user electricity usage type classification method, which is executed by a processing device of a user electricity usage type classification system, and includes the following steps:

[0013] Reading a plurality of training power consumption data sets, each of which includes a plurality of power consumption values ​​per unit time having time sequence dependencies, and each of which has power consumption characteristic information;

[0014] Using the power consumption characteristic information of the plurality of training power consumption data groups as the first-order characteristic value, and classifying the plurality of training power consumption data groups into a plurality of first-order power consumption data groups through a first machine learning clustering algorithm according to a first clustering quantity value;

[0015] Calculating at least one second-order eigenvalue of each of the plurality of training power consumption data sets by a feature extraction algorithm;

[0016] In each of the first-order power consumption data groups, according to the at least one second-order feature value of each of the training power consumption data groups and according to a second clustering quantity value of each of the first-order power consumption data groups, a second machine learning clustering algorithm is used to classify the plurality of training power consumption data groups into a plurality of second-order power consumption data groups, and a power consumption type label is assigned to each of the second-order power consumption data groups; and

[0017] A power consumption data group to be classified is received, an average similarity between the power consumption data group to be classified and the training power consumption data groups in each of the second-order power consumption data groups is calculated respectively, the power consumption type label of the second-order power consumption data group with the highest average similarity is selected as a power consumption type classification result of the power consumption data group to be classified, and the power consumption type classification result is output.

[0018] The user electricity consumption pattern classification system and method of the present invention mainly firstly performs a first grouping calculation based on a power consumption characteristic information of each electricity consumption data group as a first-order characteristic value to generate a specified number of first-order electricity consumption data groups. Then, the second-order characteristic value of the electricity consumption data group is obtained through a feature extraction algorithm, and the electricity consumption data group under each first-order electricity consumption data group is subjected to a second grouping calculation based on the second-order characteristic values ​​to generate a plurality of second-order electricity consumption data groups of a specified number, and the plurality of second-order electricity consumption data groups are given electricity consumption pattern labels.

[0019] The multiple second-order electricity usage data groups are a grouping result, and each second-order electricity usage data group represents a classification type that can reflect the user's electricity usage pattern. When the processing device receives a new electricity usage data group to be classified, the average similarity between the electricity usage data group to be classified and each of the second-order electricity usage data groups is calculated to select a group electricity usage pattern classification result in which the electricity usage pattern label is the electricity usage qualification group to be classified.

[0020] The present invention first uses the characteristic information of electricity consumption to perform the first stage of clustering, and then performs the second stage of clustering through feature extraction. It has been verified that the clustering results with obvious distinctions and different characteristics can be obtained. When the electricity consumption data group to be classified is classified according to this clustering result as a classification model, the obtained electricity consumption pattern classification result further indicates the type of electricity consumption pattern of this user on the premise of indicating the electricity consumption characteristics of the user. In this way, the electricity consumption pattern classification result can accurately reflect the actual electricity consumption pattern of the user, which is conducive to the power industry to give customized electricity adjustment suggestions, or more accurately predict the user's future electricity consumption behavior, so as to give the user reminders at the appropriate time, so that the user can adjust the electricity consumption behavior in real time. The present invention provides an accurate electricity consumption pattern classification model, which can be of great help when these electricity consumption data groups are used for electricity consumption behavior big data analysis, or further used to establish analysis and prediction models, and effectively improve the accuracy of the analysis results or the performance of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a system block diagram of the user electricity usage type classification system of the present invention.

[0022] Figure 2 It is a flow chart of the method for classifying the electricity consumption patterns of users according to the present invention.

[0023] Figure 3 It is a schematic diagram of a first-order electricity consumption data group generated by the user electricity consumption type classification system of the present invention.

[0024] Figure 4 It is a schematic diagram of the first-order and second-order electricity consumption data groups generated by the user electricity consumption type classification system of the present invention.

[0025] Figure 5It is an average curve diagram of the second-order electricity consumption data group generated by the user electricity consumption type classification system of the present invention.

[0026] Figure 6 It is an average curve diagram of another second-order electricity consumption data group generated by the user electricity consumption type classification system of the present invention. DETAILED DESCRIPTION

[0027] See also Figure 1 As shown, the user power consumption pattern classification system of the present invention mainly includes a processing device 10 and a storage device 20. The storage device 20 stores a plurality of training power consumption data sets, each of which includes a plurality of unit time power consumption values ​​with time sequence dependency, and each of which has power consumption characteristic information. The processing device 10 is connected to the storage device 20 so as to read the plurality of training power consumption data sets from the storage device 20.

[0028] Preferably, the time period of each training power consumption data set is the same, for example, one day is one time period; the unit time power consumption value in each power consumption data set is the power load per unit time interval, one unit time is, for example, 15 minutes, 30 minutes or 60 minutes, and the unit of power load is, for example, watt. For example, if the time period of each power consumption data set is one day and the sampling unit time is 30 minutes, then each power consumption data set will have 48 unit time power consumption values.

[0029] In one embodiment, when the processing device 10 receives a plurality of original power consumption data sets from a data source device, the processing device 10 first performs at least one of a data integration procedure, a data cleaning procedure, a data resampling procedure, a data standardization procedure, or a combination thereof on the plurality of original power consumption data sets to generate the plurality of training power consumption data sets, and stores the plurality of training data sets in the storage device 20 to generate training power consumption data sets that meet the required power consumption data set time period, unit time, power load unit, and other data characteristics. The data standardization procedure is, for example, a Max-min standardization procedure.

[0030] In one embodiment, the power consumption characteristic information includes a total power consumption value and a maximum power consumption value. The total power consumption value of the training power consumption data set is, for example, the sum of all the power consumption values ​​per unit time in the training power consumption data set; and the maximum power consumption value of the training power consumption data set is the highest of all the power consumption values ​​per unit time in the training power consumption data set. The total power consumption value and the maximum power consumption value of each training power consumption data set can be stored in the address corresponding to each training power consumption data set in the storage device 20 through a pre-calculation program, or can be calculated and generated when the processing device 10 reads each training power consumption data set, and the present invention is not limited to this.

[0031] See also Figure 2 As shown, the user electricity usage type classification method of the present invention is executed by the processing device 10, and includes steps S101-S105, which will be described in detail below.

[0032] In step S101 , the processing device 10 reads a plurality of training power consumption data sets from the storage device 20 .

[0033] In step S102 , the processing device 10 uses the power consumption characteristic information as the first-order characteristic value of each training power consumption data set, and classifies the plurality of training power consumption data sets into a plurality of first-order power consumption data groups through a first machine learning clustering algorithm according to a first clustering quantity value.

[0034] In one embodiment, the power consumption characteristic information includes two values, namely, the total power consumption value and the maximum power consumption value. In other words, two characteristic values ​​are used to perform the clustering calculation in the aforementioned first machine learning clustering algorithm. In this embodiment, the first-stage clustering procedure using a total of two characteristic values ​​is used, and the first-order power consumption data group generated can effectively reflect the power consumption characteristics of the multiple training power consumption data groups. This step can also be regarded as a labeling step for the multiple training power consumption data groups, as a clustering benchmark for the subsequent second-stage clustering procedure, so that the present invention as a whole can be regarded as a semi-supervised power consumption pattern clustering technology.

[0035] In other embodiments, the power consumption characteristic information may also include any numerical value that can reflect the "power consumption" characteristics of the training power consumption data group, such as the average value, median, root mean square value, etc., but the present invention is not limited thereto.

[0036] In one embodiment, the first grouping number value is 2. The first grouping number value is, for example, a user input value received through a user interface of the processing device 10, or stored in a storage module according to a user's pre-setting. In this way, a total of 2 first-order power consumption data groups are generated through the first machine learning grouping algorithm.

[0037] Please also read Figure 3 As shown, each training power consumption data group is represented by a data point D, and based on the power consumption characteristic information in step S102, the two first-order power consumption data groups generated can be regarded as corresponding to the "low power consumption group A1" and the "high power consumption group A2". In other embodiments, when the first grouping quantity value is 3, three first-order power consumption data groups will be generated, which can be regarded as representing the "first-level power consumption group", "second-level power consumption group" and "third-level power consumption group", etc., and the rest can be deduced by analogy.

[0038] The first machine learning clustering algorithm is an unsupervised clustering algorithm, such as a k-means clustering algorithm or a hierarchical clustering algorithm.

[0039] In step S103 , the processing device 10 calculates at least one second-order feature value of each of the plurality of training power usage data sets using a feature extraction algorithm.

[0040] In one embodiment, the number of at least one second-order eigenvalue is 2. In other words, the unit time power consumption values ​​of the plurality of training power consumption data sets generate 2 second-order eigenvalues ​​after the feature extraction algorithm. In one embodiment, the feature extraction algorithm is, for example, a principal component analysis (Principal Components Analysis) algorithm. The principal component analysis algorithm is mainly used to extract features from high-dimensional data to generate lower-dimensional eigenvalues ​​to improve the efficiency of subsequent operations.

[0041] Different from the feature value generation method in step S101 which uses the power consumption feature information as the first-order feature value, the second-order feature value is generated by the feature extraction algorithm to produce a more objective clustering result in the second-stage clustering procedure in the next step.

[0042] In step S104, the processing device 10 classifies the multiple training power consumption data groups into multiple second-order power consumption data groups in each first-order power consumption data group according to at least one second-order feature value of each training power consumption data group and a second clustering quantity value of each first-order power consumption data group through a second machine learning clustering algorithm, and assigns a power consumption type label to each second-order power consumption data group.

[0043] In one embodiment, the second grouping quantity value of each first-order power consumption data group is generated by a grouping quantity determination algorithm. In more detail, the processing device 10 passes the second-order characteristic value of each power consumption data group in the multiple first-order power consumption data groups through the grouping quantity determination algorithm to generate the second grouping quantity value. Taking two first-order power consumption data groups as an example, the processing device 10 first executes the grouping quantity determination algorithm according to the second-order characteristic value of the power consumption data group in the power data group A1, and generates the second grouping quantity value for the low power consumption group A1, and then executes the grouping quantity determination algorithm according to the second-order characteristic value of the power consumption data group in the high power consumption group A2, and generates the second grouping quantity value for the high power consumption group A2. In other words, even if the second-stage grouping procedure is also performed, different first-order power consumption data groups may have different numbers of second-order power consumption data groups.

[0044] Please also read Figure 4As shown, in this hypothetical example, through the grouping quantity determination algorithm, the second grouping quantity value of the low power consumption group A1 is 3, and the second grouping quantity value of the high power consumption group A2 is 2. Therefore, three second-order power consumption data groups A11~A13 are generated in the low power consumption group A1, and two second-order power consumption data groups A21 and A22 are generated in the high power consumption group A2.

[0045] This step is to further generate more refined second-order power consumption data groups based on the grouping results of the plurality of first-order power consumption data groups. The plurality of second-order power consumption data groups constitute a power consumption pattern classification model of the present invention, which can be used to classify the received power consumption data group to be classified.

[0046] The clustering number determination algorithm is the Elbow Method, and the second machine learning clustering algorithm is an unsupervised clustering algorithm, such as a k-means clustering algorithm or a hierarchical clustering algorithm.

[0047] In order to verify the effectiveness of the second-order electricity usage classification groups of the user electricity usage pattern classification system and method of the present invention, Figure 5 and Figure 6 The figure shows the result of the average curve of the second-order power consumption classification group generated by the user power consumption type classification method according to the present invention based on a set of real power consumption data as input. In this example, the second grouping number values ​​generated by the grouping number determination algorithm in the two first-order power consumption classification groups are both 4, so 4 second-order power consumption classification groups are generated in the two first-order power consumption classification groups. Among them, Figure 5 The average curves high0 to high3 of the four second-order electricity consumption classification groups under the first-order electricity consumption classification group of "high electricity consumption group" are shown in Figure 1. Figure 6 It is the average curve low0~low3 of the four second-order electricity consumption classification groups under the first-order electricity consumption classification group of "low electricity consumption group".

[0048] It should be noted that Figure 5 and Figure 6 In the charts, the scale and size of the vertical axis values ​​are different. Figure 5 In the "high power consumption group", the highest value is about 1600W, and the vertical axis (load) interval is 200W; Figure 6 In the "low power consumption group", the highest value is about 800W, and the vertical axis interval of the chart is 100W. In other words, in the first stage grouping procedure of the present invention, all power consumption data groups are effectively divided into two first-order power consumption data groups according to power consumption characteristics.

[0049] Furthermore, Figure 5 and Figure 6 Each of the eight curves, namely, curves high0 to high3 and curves low0 to low3, has a distinct shape and trend characteristics. It can be seen that the second-order power consumption data groups generated in the second-stage grouping procedure of the present invention not only have different value ranges, but also have different trend characteristics, which can represent different power consumption patterns. Therefore, each second-order power consumption data group generated by the present invention can effectively reflect different power consumption conditions.

[0050] In step S105, when the processing device 10 receives a power data set to be classified, the processing device 10 respectively calculates an average similarity between the power data set to be classified and the training power data sets in each second-order power data group, selects the power type label of the second-order power data group with the highest average similarity as a power type classification result of the power data set to be classified, and outputs the power type classification result. When the processing device 10 receives a new power data set to be classified, the power data set to be classified is classified into any one of the plurality of second-order power data groups according to the similarity calculation.

[0051] In one embodiment, when the processing device 10 calculates an average similarity between the power data group to be classified and each second-order power data group, it first calculates the average curve of the multiple training data groups in each second-order power data group, then calculates a data distance between the power data group to be classified and the average curve of each second-order power data group, and finally generates the average similarity based on an inverse value of the data distance between the power data group to be classified and each second-order power data group.

[0052] The data distance is preferably Euclidean distance. Alternatively, it can be any data distance algorithm for calculating the similarity of data values ​​in the field of machine learning technology, and the present invention is not limited thereto.

[0053] The higher the average similarity, the closer the trend between the power consumption data group to be classified and the second-order power consumption data group is. Therefore, the power consumption type label with the highest average similarity is selected as the power consumption type classification result of the power consumption data group to be classified.

[0054] In summary, the user electricity consumption pattern classification system and method of the present invention provides a complete method for generating an electricity consumption pattern classification model, as well as a complete technical solution for generating electricity consumption pattern classification results based on the electricity consumption pattern classification model. The electricity consumption pattern classification model generated by the present invention first considers the electricity consumption characteristics of the electricity consumption data group for the first stage classification, and then further performs the second stage classification based on the feature values ​​generated by the feature extraction algorithm. It has been verified that the second-order electricity consumption data group generated thereby can more accurately reflect different user electricity consumption patterns under the preliminary grouping based on electricity consumption characteristics, compared to simply classifying and analyzing the proportion of electricity consumption in different time periods, so that power companies can more accurately provide customized electricity adjustment suggestions.

[0055] In addition, since the training power consumption data set and the power consumption data set to be analyzed are based on the same time period (e.g., one day) as the data set period, the power consumption data set generated by the same user in each time period can generate power consumption pattern classification results for each time period through the present invention. After collecting data and analyzing classification results for multiple time periods, the power consumption pattern of customers on different dates can be predicted, and the power consumption habit data of users can be established. In this way, the power industry can immediately remind users when they deviate from their original power consumption habits, so that users can adjust their power consumption behavior or check for abnormalities in real time.

[0056] In addition, the more accurate electricity consumption pattern classification model of the present invention and the classification results generated therefrom can also be applied to the non-intrusive load monitoring (NILM) technology that has been actively developed in various countries in recent years, as training feature data for NILM, to improve the model performance of its monitoring, abnormality judgment, and electricity consumption prediction technologies.

[0057] Finally, the electricity consumption pattern classification model generated by the present invention can be used to analyze a large amount of electricity consumption data of a large number of electricity users, and assist power companies in formulating different rate plans, such as electricity plans that are more suitable for specific family population composition and work and rest schedules.

[0058] The above description is only an embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as above by the embodiment, it is not used to limit the present invention. Any technician familiar with the profession can make some changes or modify the technical contents disclosed above into equivalent embodiments of equivalent changes without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A user electricity usage type classification system, characterized in that: Include: A storage device stores a plurality of training power consumption data sets, each of which includes a plurality of unit time power consumption values ​​having time sequence dependencies, and each of which has power consumption characteristic information; a processing device connected to the storage device, and reading the plurality of training electrical data sets from the storage device; The processing device uses the power consumption characteristic information of the plurality of training power consumption data groups as the first-order characteristic value, and classifies the plurality of training power consumption data groups into a plurality of first-order power consumption data groups through a first machine learning clustering algorithm according to a first clustering quantity value; The processing device calculates at least one second-order eigenvalue of each of the plurality of training power consumption data sets through a feature extraction algorithm; The processing device classifies the plurality of training power consumption data groups into a plurality of second-order power consumption data groups according to the at least one second-order feature value of each training power consumption data group and a second clustering quantity value of each first-order power consumption data group through a second machine learning clustering algorithm, and assigns a power consumption type label to each second-order power consumption data group; The processing device receives a power consumption data group to be classified, calculates an average similarity between the power consumption data group to be classified and the training power consumption data groups in each of the second-order power consumption data groups, selects the power consumption type label of the second-order power consumption data group with the highest average similarity as a power consumption type classification result of the power consumption data group to be classified, and outputs the power consumption type classification result.

2. The user electricity usage type classification system as claimed in claim 1, characterized in that: The processing device generates the second clustering quantity value in the plurality of first-order power consumption data groups respectively according to the at least one second-order characteristic value of each training power consumption data group through a clustering quantity determination algorithm.

3. The user electricity usage type classification system as claimed in claim 1, characterized in that: When the device calculates an average similarity between the power data group to be classified and the training power data groups in each of the second-order power data groups, it first calculates an average curve of the multiple training data groups in each of the second-order power data groups, then calculates a data distance between the power data group to be classified and the average curve of each of the second-order power data groups, and finally generates the average similarity based on an inverse value of the data distance between the power data group to be classified and each of the second-order power data groups.

4. The user electricity usage type classification system as claimed in claim 1, characterized in that: The processing device first receives a plurality of original electricity usage data groups from a data source device, performs at least one of a data integration procedure, a data cleaning procedure, a data resampling procedure, a data standardization procedure, or a combination thereof on the plurality of original electricity usage data groups to generate the plurality of training electricity usage data groups, and stores the plurality of training electricity usage data groups in the storage device.

5. The user electricity usage type classification system as claimed in claim 2, characterized in that: The algorithm for determining the number of clusters is the elbow method.

6. The user electricity usage type classification system as claimed in claim 1, characterized in that: The first machine learning clustering algorithm and the second machine learning clustering algorithm are respectively one of a k-means clustering algorithm and a hierarchical clustering algorithm.

7. The user electricity usage type classification system as claimed in claim 1, characterized in that: The feature extraction algorithm is a principal component analysis algorithm.

8. The user electricity usage type classification system as claimed in claim 1, characterized in that: The power consumption characteristic information includes a total power consumption value and a maximum power consumption value.

9. The user electricity usage type classification system as claimed in claim 1, characterized in that: The first grouping quantity value is 2; and The number of the at least one second-order eigenvalue is 2.

10. A method for classifying user electricity usage patterns, characterized in that: The method is executed by a processing device of a user electricity usage type classification system, and includes the following steps: Reading a plurality of training power consumption data sets, each of which includes a plurality of power consumption values ​​per unit time having time sequence dependencies, and each of which has power consumption characteristic information; Using the power consumption characteristic information of the plurality of training power consumption data groups as the first-order characteristic value, and classifying the plurality of training power consumption data groups into a plurality of first-order power consumption data groups through a first machine learning clustering algorithm according to a first clustering quantity value; Calculating at least one second-order eigenvalue of each of the plurality of training power consumption data sets by a feature extraction algorithm; In each of the first-order power consumption data groups, according to the at least one second-order feature value of each of the training power consumption data groups and according to a second clustering quantity value of each of the first-order power consumption data groups, a second machine learning clustering algorithm is used to classify the plurality of training power consumption data groups into a plurality of second-order power consumption data groups, and a power consumption type label is assigned to each of the second-order power consumption data groups; as well as A power consumption data group to be classified is received, an average similarity between the power consumption data group to be classified and the training power consumption data groups in each of the second-order power consumption data groups is calculated respectively, the power consumption type label of the second-order power consumption data group with the highest average similarity is selected as a power consumption type classification result of the power consumption data group to be classified, and the power consumption type classification result is output.

11. The method for classifying user electricity usage patterns as claimed in claim 10, wherein: It also includes the following steps: In each of the plurality of first-order power usage data groups, the second clustering quantity value is generated by a clustering quantity determination algorithm according to the at least one second-order feature value of each training power usage data group.

12. The method for classifying user electricity usage patterns as claimed in claim 10, wherein: When executing "respectively calculating an average similarity between the electricity data group to be classified and the training electricity data groups in each of the second-order electricity data groups", an average curve of the multiple training data groups in each of the second-order electricity data groups is calculated first, and then a data distance between the electricity data group to be classified and the average curve of each of the second-order electricity data groups is calculated, and finally the average similarity is generated based on an inverse value of the data distance between the electricity data group to be classified and each of the second-order electricity data groups.

13. The method for classifying user electricity usage patterns as claimed in claim 10, wherein: It also includes the following steps: First, a data source device receives a plurality of original power usage data groups, and performs at least one of a data integration procedure, a data cleaning procedure, a data resampling procedure, a data standardization procedure, or a combination thereof on the plurality of original power usage data groups to generate the plurality of training power usage data groups, and the plurality of training power usage data groups are stored in the storage device.

14. The method for classifying user electricity usage patterns as claimed in claim 11, wherein: The algorithm for determining the number of clusters is the elbow method.

15. The method for classifying user electricity usage patterns as claimed in claim 10, wherein: The first machine learning clustering algorithm and the second machine learning clustering algorithm are respectively one of a k-means clustering algorithm and a hierarchical clustering algorithm.

16. The method for classifying user electricity usage patterns as claimed in claim 10, wherein: The feature extraction algorithm is a principal component analysis algorithm.

17. The method for classifying user electricity usage patterns as claimed in claim 10, wherein: The power consumption characteristic information includes a total power consumption value and a maximum power consumption value.

18. The method for classifying user electricity usage patterns as claimed in claim 10, wherein: The first grouping quantity value is 2; and The number of the at least one second-order eigenvalue is 2.

Images