CN111177216B - Association rule generation method and device for comprehensive energy consumer behavior characteristics - Google Patents

Abstract

The invention relates to a method for generating association rules for synthesizing behavior characteristics of energy consumers, which comprises the following steps: step 1, carrying out normalization processing on time sequence data of an intelligent ammeter; step 2, converting the time sequence data of the intelligent ammeter into symbolic representation; step 3, extracting characteristic patterns in the symbol and adding characteristic motifs of the characteristic patterns into a subject library; step 4, carrying out time association rule mining on the feature motifs of the newly added feature patterns in the theme library, and analyzing the relation between influence factors which cause energy consumption change in a certain specific period; step 5, performing clustering data analysis on characteristic motifs in the subject database by adopting a K-means clustering method, a hierarchical method and a density clustering algorithm respectively, and generating a daily consumption profile; and 6, measuring the fitting degree of the daily consumption profile group created by the three clustering methods and the actual daily consumption situation. The invention can accurately describe the consumption condition of the real energy.

Description

Association rule generation method and device integrating behavioral characteristics of energy consumers

Technical field

The invention belongs to the field of smart meter data mining technology, relates to user electricity consumption information, and in particular, a method and device for generating association rules that integrate behavioral characteristics of energy consumers.

Background technique

Smart grid is one of the promising technologies to meet the growing energy demand and reduce global environmental pollution. It improves the efficiency, reliability, sustainability and economy of electrical energy. Over the past decade, smart meters have been deployed in much of the world. Smart meters and database management systems form Advanced Metering Infrastructure (AMI), which plays an important role in energy systems by facilitating two-way information flow and recording energy distribution. AMI has produced various novel smart home services such as energy saving and awareness recommendations to end users. Smart meters have great potential to analyze fine-grained energy consumption data for energy planning and management. The deployment of smart energy meters benefits both energy consumers and utility professionals.

Time series data generated by smart meters has great potential to identify routine and abnormal energy consumption patterns. Time series data mining techniques are modeled and developed to identify the energy consumption behavior of energy consumers. Smart meter data requires advanced data analytics for accurate and automated decision making in a real-time environment. Through dynamic pricing, it can increase consumer energy awareness by providing a better understanding of how and when energy is used. Energy data analysis has become a major research area in electricity consumption analysis. The ability to analyze smart meter data to identify day-to-day activity can be useful for power companies implementing demand-side management techniques.

In smart grids, renewable energy sources are increasingly becoming more prevalent. However, the intermittent nature of renewable energy generation has led to the contradiction between supply and demand. Therefore, dynamic energy trading prices throughout the day make the time aspect even more important. Smart meter energy consumption patterns can fluctuate differently throughout the day depending on the time or month, weather, occupant schedules and behavior. Likewise, grid load changes temporally as demand, temperature and renewable generation change, which are in turn influenced by weather and seasonal timescales.

In recent years, various techniques have been developed to mine time series data. However, the temporal nature of time series energy consumption data has only been studied to a limited extent, but since energy consumption is a highly dynamic concept with different load demands and pricing over time. Therefore, in order to accurately describe real energy consumption, an association rule generation method and device are needed that has a short response time and can achieve comprehensive energy consumer behavior characteristics that are frequently sampled within a period of time.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings of the existing technology and provide an association rule generation method and device with reasonable design, short response time and the ability to achieve frequent sampling of comprehensive energy consumer behavior characteristics within a period of time.

The present invention solves its technical problems by adopting the following technical solutions:

An association rule generation method that integrates behavioral characteristics of energy consumers, including the following steps:

Step 1. Normalize the smart meter time series data;

Step 2. Use symbolic approximation clustering to first perform cloud segmentation aggregation approximation on the normalized smart meter time series data, and then convert it into symbolic representation;

Step 3. Extract the feature pattern in the symbolic representation result and add the feature motif of the feature pattern to the topic library, where the feature motif satisfies the user-defined frequency count and availability threshold;

Step 4. Execute time association rule mining on the feature motifs of the newly added feature patterns in the theme library, and analyze the connection between the influencing factors that lead to changes in energy consumption within a specific period. If the correlation between the influencing factors that lead to changes in energy consumption is If the absolute value of the coefficient is greater than the set value, proceed to the next step; otherwise, return to step 3 to re-extract the feature pattern and add the feature motif of the feature pattern to the theme library;

Step 5: After analyzing the clustering data of the feature motifs in the subject database using three clustering methods: K-means clustering, hierarchical method and density clustering algorithm, execute the feature patterns corresponding to the feature motifs. The clustering results are used to generate daily consumption profiles;

Step 6: Statistically evaluate the daily consumption profiles generated by the three clustering methods through the two metrics of the sum of squared errors and the silhouette coefficient to measure the difference between the daily consumption profile groups created by the three clustering methods and the actual The fitting degree of daily consumption situation.

Moreover, the specific method of step 1 is:

Normalize the data to unit variance using z normalization, where the normalized value The mean μ and standard deviation σ are as follows:

In the above formula, _xi is the data to be processed, and n is the number of data to be processed.

Moreover, the specific steps of step 2 include:

(1) Convert time series data into cloud segmented aggregation approximate representation symbols;

(2) Convert cloud segment aggregation approximate representation symbols into strings.

Moreover, the specific steps of step 2 (1) include:

①Represent the current segmented sequence data in a cloud model;

②Use the entropy of each cloud model to evaluate the data stability of the subsequence, and select the subsequence with the worst stability (the smallest entropy) (denoted as Q(i ₀ , j ₀ )) for segmented aggregation;

③Find a data point as a key point q _k in the segmented aggregation subsequence Q(i ₀ , j ₀ ), i ₀ <k <j ₀ , this key point q _k can make the two subsequences separated by it The difference between the sum of cloud model entropy of (Q(i ₀ , k) and Q(k, j ₀ )) and the cloud model entropy of subsequence Q(i ₀ , j ₀ ) is the largest, and subsequence Q is deleted at the same time. (i ₀ , j ₀ ), record subsequences Q(i ₀ , k) and Q(k, j ₀ );

④ Repeat ①～③ until the stop condition is met.

Moreover, the specific method of step (2) of step 2 is:

After the cloud segmentation aggregation approximate representation, the discretized representation is symbolized, converting the time series into a symbolic string.

Moreover, the specific steps of step 3 include:

(1) After symbolic approximate clustering conversion, SAX pattern types are generated; and the pattern types generated by SAX are marked as topics or uncommon patterns respectively; where topics are defined as previously unknown, frequently occurring patterns;

(2) Extract the required feature pattern from the data transformed by symbolic approximate clustering, and add the feature motif of the feature pattern to the subject library.

Moreover, the specific method of step 4 is:

Among them, X is what happened first, Y is what happened later, and N is the total number of records or transactions rules; the confidence of score.

Moreover, the specific steps of step 5 include:

(1) Use K-means clustering method to cluster the characteristic motifs in the topic library:

Using the Euclidean distance metric, daily blocks (N ₁ , N ₂ ,..., N ₃ ) are divided into k sets, S = (S ₁ , S ₂ ...S _k ), to obtain the minimized sum of squares :

where μ _i is the average of the values in Si _i ;

(2) Use the hierarchical method to cluster the feature motifs in the topic library:

A bottom-up hierarchical algorithm is used for clustering, and then an iterative relocation method is used to improve the clustering results obtained by the hierarchical algorithm;

(3) Use density clustering algorithm to cluster the feature motifs in the topic library:

①Set the values of scanning radius eps and minimum number of included points minPts;

② Select an unvisited point and find all nearby points within the scanning radius eps;

③If the number of nearby points ≥ the minimum number of included points minPts, the current point and its nearby points form a cluster, and the starting point is marked as visited; then recursively, use the same method to process all points in the cluster that are not marked as visited , thereby expanding the cluster; if the number of nearby points < the minimum number of included points minPts, the point is temporarily marked as a noise point;

④If the cluster in step ③ is fully expanded, that is, all points in the cluster are marked as visited, then the same algorithm is used to process unvisited points.

Moreover, the specific steps of step 6 include:

(1) The sum of squared errors SSE is a measure of the tightness of the cluster. The smaller the SSE, the better the quality of the cluster. Its calculation formula is:

Among them, (N ₁ , N ₂ , ..., N ₃ ) are daily data groups, and these data are divided into k sets S = (S ₁ , S ₂ ... S _k ).

(2) Silhouette coefficient is a measure of inter-cluster cohesion and intra-cluster separation. The coefficient 1 is the best and -1 is the worst. The calculation formula of the silhouette coefficient is:

where a is the average distance between a data instance and all other points in the same cluster, and b is the average distance between a data instance and all other points in the cluster closest to it.

An association rule generation device that integrates behavioral characteristics of energy consumers, including:

Normalization processing module, used to normalize smart meter time series data;

The symbolic conversion module is used to use symbolic approximate clustering to first perform cloud segmentation aggregation approximation on the normalized smart meter time series data, and then convert it into symbolic representation;

A feature pattern extraction module for extracting feature patterns in symbols and adding the feature motifs of the feature patterns to the theme library, meeting user-defined frequency counts and availability thresholds;

The time association rule mining module is used to mine the feature motifs of newly added feature patterns in the theme library and analyze the connection between the influencing factors that lead to changes in energy consumption within a specific period. If the changes in energy consumption are caused If the connection between the influencing factors is strong, proceed to the next step; if the connection is weak, return to the feature pattern extraction module to re-extract the feature pattern and add the feature motif of the feature pattern to the theme library;

The clustering module uses three clustering methods: K-means clustering, hierarchical method and density clustering algorithm to analyze the clustering data of the feature motifs in the subject library, and then executes the features corresponding to the feature motifs. Clustering results of patterns to generate daily consumption profiles;

The clustering result evaluation module statistically evaluates the daily consumption profiles generated by the three clustering methods through the two metrics of the sum of squared errors and the silhouette coefficient to measure the daily consumption profiles created by the three clustering methods. How well the group fits actual daily consumption.

The advantages and positive effects of the present invention are:

1. The present invention proposes a method and device for generating association rules that integrate behavioral characteristics of energy consumers. The method mines association rules for smart meter data within a specific time window, thereby characterizing the electricity consumption behavior of smart meter users. The present invention first symbolically processes smart meter data to facilitate the application of various data mining technologies; secondly, based on primitive theme recognition, it can help identify the energy relationship of consumer behavior patterns and extract characteristic patterns to satisfy users. Defined frequency counts and availability thresholds; third, conduct temporal association rule mining to analyze the connections between certain factors that may cause an increase/decrease in energy consumption within a specific period; finally, perform pattern-based clustering to create energy Daily overview of consumption. The present invention can accurately describe real energy consumption.

2. The present invention converts energy consumption values into symbolic representation, greatly reducing the dimensionality of the time series. The present invention uses symbolic representation that can be used both locally on embedded sensing systems for home automation and by practical experts to efficiently process smart meter data.

3. The primitives of the present invention are defined as previously unknown and frequently occurring patterns in time series. It can discover similarities that represent the same transactions in real life. The patterns and their time tags can be used to identify families at a specific time. Behavior, therefore the application of primitives can play an important role in understanding household behavior and determining household patterns, and the time information extracted by this method can accurately describe real energy consumption.

Description of drawings

Figure 1 is a flow chart of the steps of the present invention

Figure 2 is a processing flow chart of the present invention;

Figure 3 is a processing flow chart of the cloud segment aggregation approximate representation symbol of the present invention;

Figure 4 is a processing flow chart of the density clustering algorithm of the present invention.

Detailed ways

The embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings:

The association rule generation method for comprehensive energy consumer behavior characteristics is shown in Figure 1, Figure 2, Figure 3 and Figure 4, which includes the following steps:

Step 1. Normalize the smart meter time series data;

The specific method of step 1 is:

Normalize the data to unit variance with z-normalization, where the normalized value x, mean μ, and standard deviation σ are as follows:

Step 2. Use symbolic approximate clustering to convert the preprocessed smart meter time series data into symbolic representation;

In step 2, Symbolic aggregate approximation (SAX) is used to convert the preprocessed time series data into symbolic representation. Using SAX can effectively achieve dimensionality reduction and prepare the data in a format suitable for applying different data mining techniques.

The specific steps of step 2 include:

(1) Convert time series data into cloud segmented aggregation approximate representation symbols;

The z-normalized time series data is discretized using cloud piecewise aggregation approximation C (PAA). In this representation, the entropy of each cloud model is used to evaluate the data stability of the subsequences, and the cloud segmentation aggregation approximation is re-performed for the subsequences that do not meet the requirements, thereby dividing them into w-dimensional space.

The specific steps of step 2 (1) include:

①Represent the current segmented sequence data in a cloud model;

②Use the entropy of each cloud model to evaluate the data stability of the subsequence, and select the subsequence with the worst stability (the smallest entropy) (denoted as Q(i ₀ , j ₀ )) for segmented aggregation;

③Find a data point as a key point q _k in the segmented aggregation subsequence Q(i ₀ , j ₀ ), i ₀ <k <j ₀ , this key point q _k can make the two subsequences separated by it The difference between the sum of cloud model entropy of (Q(i ₀ , k) and Q(k, j ₀ )) and the cloud model entropy of subsequence Q(i ₀ , j ₀ ) is the largest, and subsequence Q is deleted at the same time. (i ₀ , j ₀ ), record subsequences Q(i ₀ , k) and Q(k, j ₀ );

④ Repeat ①～③ until the stop condition is met.

(2) Convert cloud segment aggregation approximate representation symbols into strings;

The specific method of step 2 (2) is:

After the cloud segmentation aggregation approximate representation, the discretized representation is symbolized, converting the time series into a symbolic string.

In this embodiment, English letters are used to represent the original sequence. Generally speaking, the symbol "a" represents low energy consumption, "b" represents average value, "c" represents higher than average value, and "d" represents high energy consumption. After SAX transformation, descriptive knowledge types (e.g. topics, association rule mining) in time series data can be applied to knowledge discovery.

Step 3. Extract feature patterns and add feature motifs of feature patterns to the subject library that meet user-defined frequency counts and availability thresholds.

The specific steps of step 3 include:

(1) After symbolic approximate clustering conversion, SAX pattern types are generated; and the pattern types generated by SAX are marked as topics or uncommon patterns respectively; where topics are defined as previously unknown, frequently occurring patterns;

(2) Extract the required feature pattern from the data transformed by symbolic approximate clustering, and add the feature motif of the feature pattern to the subject library;

In this example, we focus on the type of patterns generated by SAX after the symbolic approximate clustering SAX transformation.

Mark SAX-generated pattern types as themes or uncommon patterns, with themes being defined as previously unknown, frequently occurring patterns.

Alphabet size A should be fixed as a reasonable compromise, since having too many symbols will produce too many patterns that may not be repeated, on the other hand having few symbols will not capture more consumer resolution. The number of windows per day W must also be chosen carefully.

A large number of patterns can be extracted from the data. It is not necessary to use all discovered patterns for analysis. The frequency and availability of patterns play an important role in detecting the regular behavior of smart meters. For example, the most common and second most common themes are important to the analysis. We can set a threshold for selecting feature patterns. Feature patterns are patterns that meet different criteria, such as the number of occurrences in a specific period, exceeding a threshold. The threshold can be set as a fraction of the frequency of each motif versus the total number of all motifs.

Additionally, generally speaking, an electrical expert with some expertise and experience will test all topics to determine which ones are interesting or not. Long periods of inactivity, other than a change in meter reading, will result in a series of similar patterns. For example, when using a letter size of 5, in addition to a small drop in energy consumption, long periods of no activity or events will result in patterns such as ccccca, ccccac, cccacc, etc. Because only one of these is interesting. Further analysis, excluding other topics, starts with two or more c's. If a pattern represents only an increase in energy consumption, it is considered meaningless. The patterns found should reflect the complete behavior, and patterns showing increased energy expenditure only represent the beginning of an event or activity. A useful topic would include start activities (turning the device on) and completion activities (turning the device off). Finally, the time of pattern and its availability within the day, week and month are important for different energy consumption behaviours.

In this way, feature motifs are extracted and added to the topic library for further knowledge discovery. These patterns have the potential to determine the energy consumption behavior of each consumer.

Step 4. Execute time association rule mining on the feature motifs of the newly added feature patterns in the theme library, and analyze the connection between the influencing factors that lead to changes in energy consumption within a specific period. If the impact of changes in energy consumption is If the absolute value of the correlation coefficient between factors is greater than the set value 0.8, proceed to the next step; otherwise, return to step 3 to re-extract the feature pattern and add the feature motif of the feature pattern to the theme library;

The specific method of step 4 is:

Among them, X is what happened first, Y is what happened later, and N is the total number of records or transactions. The confidence level of Fraction.

In this embodiment, association rule mining discovers cross-sectional associations without considering temporal information. Since electricity has complex dynamics, temporal association rule mining is a hot topic in our research. Temporal association rule mining finds relationships between variables within a specific time period.

The proposed association rule mining method is based on a motif library containing characteristic motif information for a specific time period. Fetching of frequent topics must be done from repositories whose support count is greater than or equal to the user-supplied minimum support threshold. For example, if X is the antecedent and Y is the subsequent result, the association rule X→Y means that if X occurs, Y will also occur. The rule is supported by the ratio of the number of antecedent and subsequent occurrences to the total number of records. Support for association rules indicates statistical significance. Association rules with lower support indicate those relationships that are uncommon, and rules with higher support describe those relationships that are common in records.

Where X is what happened first, Y is what happened later, and N is the total number of records or transactions. The confidence of the rule . Confidence indicates statistical strength. A higher confidence indicates a stronger correlation between the antecedent and subsequent words, while a lower confidence indicates a weak correlation.

In power dynamics, association rules that occur within a specific time period are of particular interest to us. The format of this rule is It indicates that Y will occur within the Ti time slot after X occurs. The connection between appliances can be explained by the shape of electricity usage.

Step 5. Use three clustering methods: K-means clustering, hierarchical method and density clustering algorithm to cluster the feature motifs in the topic database. After analyzing the clustering data, perform clustering based on feature patterns. Class to generate daily consumption profiles;

In this embodiment, after the motifs are discovered, the characteristic motifs are clustered to generate a daily consumption profile. The daily profile represents the household electricity consumption pattern. This is an important step if there are 15 characteristic patterns and power experts need to further aggregate these patterns into 5 or 6 typical energy consumption scenarios. It will provide the user with additional controls to further describe the properties that can be used to select parameter A or W during SAX conversion. There are various existing clustering approaches targeting different application goals. Time series clustering is a feature-based model clustering method

The specific steps of step 5 include:

(1) Due to the simplicity of K-means clustering, we use the Euclidean distance metric. The daily blocks (N ₁ , N ₂ , ..., N ₃ ) are divided into k sets, S = (S ₁ , S ₂ ..., S _k ) in order to minimize the sum of squares.

where μ _i is the average of the values in _Si

(2) Balanced iterative reduction and clustering using hierarchical methods

Balanced Iterative Reducing and Clustering Using Hierarchies (BIRCH) using hierarchical methods is a very effective and traditional hierarchical clustering algorithm. In a given motif library, the BIRCH algorithm can efficiently perform clustering in one pass and can effectively handle outliers.

The algorithm first uses a bottom-up hierarchical algorithm and then uses iterative relocation to improve the results. Hierarchical agglomeration adopts a bottom-up strategy, first treating each object as an atomic cluster, and then merging these atomic clusters to form larger clusters, reducing the number of clusters until all objects are in one cluster, or a certain terminal condition satisfied.

(3) Density clustering algorithm for clustering

Density clustering algorithm, which defines a cluster as the largest set of density-connected points, can divide areas with sufficiently high density into clusters and can discover clusters of arbitrary shapes in noisy spatial databases. Density clustering calculation requires two parameters: scanning radius (eps) and minimum number of included points (minPts). Start from an unvisited point and find all nearby points within eps (including eps).

If the number of nearby points ≥ minPts, the current point and its nearby points form a cluster, and the starting point is marked as visited. Then recurse and process all points in the cluster that are not marked as visited in the same way to expand the cluster.

If the number of nearby points < minPts, the point is temporarily marked as a noise point.

If the cluster is sufficiently expanded, that is, all points in the cluster are marked as visited, then the same algorithm is used to process unvisited points.

Step 6. Statistically evaluate the three clustering methods through two metrics: the sum of squared errors and the silhouette coefficient, respectively, to measure the extent to which k-means clustering can create daily consumption profile groups.

The specific steps of step 6 include:

(1) The sum of squared errors SSE is a measure of the tightness of the cluster. Smaller errors indicate good cluster quality. Its calculation formula is:

Among them, (N ₁ , N ₂ ,..., N ₃ ) are daily data groups, and these data are divided into k sets S=(S ₁ , S ₂ ...S _k ).

(2) Silhouette coefficient is a measure of inter-cluster cohesion and intra-cluster separation. The coefficient 1 is the best and -1 is the worst. The calculation formula of the silhouette coefficient is:

where a is the average distance between a data instance and all other points in the same cluster, and b is the average distance between a data instance and all other points in the cluster closest to it.

For instructions that the software invention can be stored in a computer-readable storage medium, please save the following template: Those skilled in the art should understand that embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment that combines software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each process and/or block in the flowchart illustrations and/or block diagrams, and combinations of processes and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce a use A device for realizing the functions specified in one process or multiple processes of the flowchart and/or one block or multiple blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory that causes a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction means, the instructions The device implements the functions specified in a process or processes of the flowchart and/or a block or blocks of the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device, causing a series of operating steps to be performed on the computer or other programmable device to produce computer-implemented processing, thereby executing on the computer or other programmable device. Instructions provide steps for implementing the functions specified in a process or processes of a flowchart diagram and/or a block or blocks of a block diagram. Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that the present invention can still be modified. Modifications or equivalent substitutions may be made to the specific embodiments, and any modifications or equivalent substitutions that do not depart from the spirit and scope of the invention shall be covered by the scope of the claims of the invention.

Claims (9)

1. An association rule generation method that integrates energy consumer behavior characteristics, which is characterized by: including the following steps: Step 1. Normalize the smart meter time series data; Step 2. Use symbolic approximation clustering to first perform cloud segmentation aggregation approximation on the normalized smart meter time series data, and then convert it into symbolic representation; Step 3. Extract the feature pattern in the symbolic representation result and add the feature motif of the feature pattern to the topic library, where the feature motif satisfies the user-defined frequency count and availability threshold; Step 4. Execute time association rule mining on the feature motifs of the newly added feature patterns in the theme library, and analyze the connection between the influencing factors that lead to changes in energy consumption within a specific period. If the correlation between the influencing factors that lead to changes in energy consumption is If the absolute value of the coefficient is greater than the set value, proceed to the next step; otherwise, return to step 3 to re-extract the feature pattern and add the feature motif of the feature pattern to the theme library; Step 5: After analyzing the clustering data of the feature motifs in the subject database using three clustering methods: K-means clustering, hierarchical method and density clustering algorithm, execute the feature patterns corresponding to the feature motifs. The clustering results are used to generate daily consumption profiles; Step 6: Statistically evaluate the daily consumption profiles generated by the three clustering methods through the two metrics of the sum of squared errors and the silhouette coefficient to measure the difference between the daily consumption profile groups created by the three clustering methods and the actual The degree of fitting of the daily consumption situation; The specific steps of step 2 include: (1) Convert time series data into cloud segmented aggregation approximate representation symbols; (2) Convert cloud segment aggregation approximate representation symbols into strings. 2. A method for generating association rules that integrates behavioral characteristics of energy consumers according to claim 1, characterized in that: the specific method of step 1 is: Normalize the data to unit variance using z normalization, where the normalized value The mean μ and standard deviation σ are as follows: In the above formula, _xi is the data to be processed, and n is the number of data to be processed. 3. A method for generating association rules that integrates behavioral characteristics of energy consumers according to claim 1, characterized in that: the specific steps of step 2 (1) include: ①Represent the current segmented sequence data in a cloud model; ②Use the entropy of each cloud model to evaluate the data stability of the subsequence, and select the subsequence with the worst stability (the smallest entropy) (denoted as Q(i ₀ , j ₀ )) for segmented aggregation; ③Find a data point as a key point q _k in the segmented aggregation subsequence Q(i ₀ , j ₀ ), i ₀ <k <j ₀ , this key point q _k can make the two subsequences separated by it The difference between the sum of cloud model entropy of (Q(i ₀ , k) and Q(k, j ₀ )) and the cloud model entropy of subsequence Q(i ₀ , j ₀ ) is the largest, and subsequence Q is deleted at the same time. (i ₀ , j ₀ ), record subsequences Q(i ₀ , k) and Q(k, j ₀ ); ④ Repeat ①～③ until the stop condition is met. 4. A method for generating association rules that integrates behavioral characteristics of energy consumers according to claim 1, characterized in that: the specific method of step (2) of step 2 is: After the cloud segmentation aggregation approximate representation, the discretized representation is symbolized, converting the time series into a symbolic string. 5. A method for generating association rules that integrates behavioral characteristics of energy consumers according to claim 1, characterized in that: the specific steps of step 3 include: (1) After symbolic approximate clustering conversion, SAX pattern types are generated; and the pattern types generated by SAX are marked as topics or uncommon patterns respectively; where topics are defined as previously unknown, frequently occurring patterns; (2) Extract the required feature pattern from the data transformed by symbolic approximate clustering, and add the feature motif of the feature pattern to the subject library. 6. A method for generating association rules that integrates behavioral characteristics of energy consumers according to claim 1, characterized in that: the specific method of step 4 is: Among them, X is what happened first, Y is what happened later, and N is the total number of records or transactions. The confidence level of Fraction. 7. A method for generating association rules that integrates behavioral characteristics of energy consumers according to claim 1, characterized in that: the specific steps of step 5 include: (1) Use K-means clustering method to cluster the characteristic motifs in the topic library: Using the Euclidean distance metric, daily blocks (N ₁ , N ₂ ,..., N ₃ ) are divided into k sets, S = (S ₁ , S ₂ ...S _k ), to obtain the minimized sum of squares : where μ _i is the average of the values in Si _i ; (2) Use the hierarchical method to cluster the feature motifs in the topic library: A bottom-up hierarchical algorithm is used for clustering, and then an iterative relocation method is used to improve the clustering results obtained by the hierarchical algorithm; (3) Use density clustering algorithm to cluster the feature motifs in the topic library: ①Set the values of scanning radius eps and minimum number of included points minPts; ② Select an unvisited point and find all nearby points within the scanning radius eps; ③If the number of nearby points ≥ the minimum number of included points minPts, the current point and its nearby points form a cluster, and the starting point is marked as visited; then recursively, use the same method to process all points in the cluster that are not marked as visited , thereby expanding the cluster; if the number of nearby points < the minimum number of included points minPts, the point is temporarily marked as a noise point; ④If the cluster in step ③ is fully expanded, that is, all points in the cluster are marked as visited, then the same algorithm is used to process unvisited points. 8. A method for generating association rules that integrates behavioral characteristics of energy consumers according to claim 1, characterized in that: the specific steps of step 6 include: (1) The sum of squared errors SSE is a measure of the tightness of the cluster. The smaller the SSE, the better the quality of the cluster. Its calculation formula is: Among them, (N ₁ , N ₂ ,..., N ₃ ) are daily data groups, and these data are divided into k sets S = (S ₁ , S ₂ ...S _k ); (2) Silhouette coefficient is a measure of inter-cluster cohesion and intra-cluster separation. The coefficient 1 is the best and -1 is the worst. The calculation formula of the silhouette coefficient is: where a is the average distance between a data instance and all other points in the same cluster, and b is the average distance between a data instance and all other points in the cluster closest to it. 9. An association rule generation device that integrates behavioral characteristics of energy consumers, including: Normalization processing module, used to normalize smart meter time series data; The symbolic conversion module is used to use symbolic approximate clustering to first perform cloud segmentation aggregation approximation on the normalized smart meter time series data, and then convert it into symbolic representation; A feature pattern extraction module for extracting feature patterns in symbols and adding the feature motifs of the feature patterns to the theme library, meeting user-defined frequency counts and availability thresholds; The time association rule mining module is used to mine the feature motifs of newly added feature patterns in the theme library and analyze the connection between the influencing factors that lead to changes in energy consumption within a specific period. If the changes in energy consumption are caused If the connection between the influencing factors is strong, proceed to the next step; if the connection is weak, return to the feature pattern extraction module to re-extract the feature pattern and add the feature motif of the feature pattern to the theme library; The clustering module uses three clustering methods: K-means clustering, hierarchical method and density clustering algorithm to analyze the clustering data of the feature motifs in the subject library, and then executes the features corresponding to the feature motifs. Clustering results of patterns to generate daily consumption profiles; The clustering result evaluation module statistically evaluates the daily consumption profiles generated by the three clustering methods through the two metrics of the sum of squared errors and the silhouette coefficient to measure the daily consumption profiles created by the three clustering methods. How well the group fits actual daily consumption.