A concentrator intelligent fusion terminal based on user behavior analysis and regulation

Abstract

The present invention relates to the field of data processing technology, and specifically to a concentrator intelligent fusion terminal based on user behavior analysis and control; obtaining feature weights based on the distribution characteristics and preset importance of power consumption feature values; performing hierarchical clustering based on the power consumption feature values ​​and feature weights; obtaining a feature difference degree based on the discrete characteristics of the power consumption feature values ​​in any parent node of the clustering tree and the difference characteristics of the power consumption feature values ​​between users of the corresponding two child nodes; obtaining a clustering error degree based on the difference characteristics of the clustering effects of any parent node and the corresponding two child nodes; and obtaining a distance difference degree based on the difference characteristics of the clustering distances between any parent node and the corresponding two child nodes. The present invention obtains abnormal feature values ​​based on the feature difference degree, clustering error degree, and distance difference degree; performs pruning based on the abnormal feature values ​​of the parent node to obtain a final user clustering result, thereby improving the accuracy of clustering and power consumption control.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a concentrator intelligent fusion terminal based on user behavior analysis and regulation. Background Art

[0002] With the development of smart grids, concentrators, as key devices connecting smart meters to master station systems, are gradually becoming intelligent. Concentrators collect user electricity usage information and cluster users based on this information, classifying them according to their different electricity usage habits and thus enabling differentiated regulation of their electricity usage. Existing hierarchical clustering algorithms can cluster users with different electricity usage behaviors. During the clustering process, clusters are combined and a cluster tree is generated by comparing the distances between clusters. However, due to the complexity of user electricity usage behavior and the diversity of electricity usage data types, clustering based on the distance between electricity usage features can easily result in users with different electricity usage behaviors being grouped in the same cluster. Once the clustering is completed, the clustering cannot be undone, making the final clustering results inaccurate and unreasonable, reducing the accuracy of user clustering and resulting in low accuracy in electricity behavior analysis and regulation. Summary of the Invention

[0003] To address the technical problem that the hierarchical clustering algorithm is used to classify users with different electricity consumption habits, resulting in inaccurate clustering results and affecting the accuracy of electricity consumption behavior analysis and electricity consumption regulation, the present invention aims to provide a concentrator intelligent fusion terminal based on user behavior analysis and regulation. The technical solution adopted is as follows:

[0004] A data acquisition module is used to obtain different types of electricity consumption characteristic values ​​of users;

[0005] An electricity consumption feature analysis module is configured to obtain feature weights based on the distribution characteristics and preset importance of the same type of electricity consumption feature values ​​of all users; hierarchical clustering is performed on all users based on the electricity consumption feature values ​​and the feature weights to obtain a clustering tree;

[0006] A cluster analysis module is configured to obtain a characteristic difference degree based on the discrete characteristics of the power consumption characteristic values ​​of users in any parent node of the clustering tree and the difference characteristics of the power consumption characteristic values ​​between users of two child nodes corresponding to the arbitrary parent node; obtain a clustering error degree of the arbitrary parent node based on the difference characteristics of the clustering effects of the arbitrary parent node and the corresponding two child nodes; and obtain a distance difference degree of the arbitrary parent node based on the difference characteristics of the clustering distances between the arbitrary parent node and the corresponding two child nodes.

[0007] The clustering adjustment module is used to obtain the abnormal feature value of any parent node according to the feature difference degree, the clustering error degree and the distance difference degree; prune the abnormal feature values ​​of all parent nodes in the clustering tree to obtain the final user clustering result.

[0008] Furthermore, the step of obtaining feature weights according to the distribution characteristics and preset importance of the same type of power consumption feature values ​​of all users includes:

[0009] The entropy weight corresponding to each type of electricity consumption characteristic value of all users is calculated by the entropy weight method; the product of a preset first proportion value and the entropy weight of any type of electricity consumption characteristic value is calculated to obtain a first value; the product of a preset second proportion value and the preset importance corresponding to any type of electricity consumption characteristic value is calculated to obtain a second value; the sum of the first value and the second value is calculated to obtain the characteristic weight of the said any type of electricity consumption characteristic value.

[0010] Furthermore, the step of performing hierarchical clustering on all users according to the electricity consumption characteristic values ​​and the characteristic weights to obtain a clustering tree includes:

[0011] The weighted distance is obtained by multiplying the difference distance of the same type of electricity consumption feature values ​​between any two users by the corresponding feature weight. The combined distance between any two users is calculated to obtain the comprehensive distance between the two users. Hierarchical clustering is performed based on the comprehensive difference distances between all users to obtain a clustering tree after user clustering is completed.

[0012] Furthermore, the step of obtaining the degree of feature difference based on the discrete features of the power consumption feature values ​​of the users in any parent node of the clustering tree and the difference features of the power consumption feature values ​​between the users of two child nodes corresponding to the arbitrary parent node includes:

[0013] Calculate the standard deviation of the same type of electricity consumption characteristic values ​​of users in the arbitrary parent node to obtain the degree of dispersion; obtain the average characteristic value based on the average value of the same type of electricity consumption characteristic values ​​of users in any child node corresponding to the arbitrary parent node; calculate the absolute value of the difference between the same type of average characteristic values ​​of the two child nodes corresponding to the arbitrary parent node; obtain the characteristic difference value; calculate the ratio of the characteristic difference value to the degree of dispersion to obtain a third value; calculate the sum of the third values ​​of all types and positively map them to obtain the degree of characteristic difference of the arbitrary parent node.

[0014] Furthermore, the step of obtaining the clustering error degree of the arbitrary parent node according to the difference characteristics of the clustering effects of the arbitrary parent node and the corresponding two child nodes includes:

[0015] Calculate the average value of the silhouette coefficients of the two child nodes corresponding to the arbitrary parent node; obtain the average silhouette coefficient; calculate the difference between the silhouette coefficient of the arbitrary parent node and the average silhouette coefficient and perform negative correlation mapping to obtain the clustering error degree of the arbitrary parent node.

[0016] Furthermore, the step of obtaining the distance difference degree of the arbitrary parent node according to the difference characteristics of the cluster distances between the arbitrary parent node and the corresponding two child nodes includes:

[0017] The average value of the clustering distances when generating the two child nodes corresponding to the arbitrary parent node is calculated to obtain the average clustering distance; the difference between the clustering distance when generating the arbitrary parent node and the average clustering distance is calculated and positively correlated with each other to obtain the distance difference degree of the arbitrary parent node.

[0018] Furthermore, the step of obtaining the abnormal feature value of the arbitrary parent node according to the feature difference degree, the clustering error degree and the distance difference degree includes:

[0019] The average values ​​of the feature difference degree, the clustering error degree, and the distance difference degree are calculated to obtain the abnormal feature value of the arbitrary parent node.

[0020] Furthermore, the step of pruning according to the abnormal feature values ​​of all parent nodes in the clustering tree to obtain the final user clustering result includes:

[0021] When the abnormal feature value of any parent node in the clustering tree exceeds a preset abnormal threshold, the arbitrary parent node is pruned; and the user distribution result in the clustering tree after pruning is used as the final user clustering result.

[0022] Furthermore, the electricity consumption characteristic values ​​include: average daily electricity consumption, coefficient of variation of daily electricity consumption, peak-to-valley electricity ratio, and maximum load.

[0023] The present invention has the following beneficial effects:

[0024] In the present invention, since different types of electricity usage features contribute differently to the similarity of electricity usage habits between users, obtaining feature weights can determine the importance of different types of electricity usage features in clustering user electricity usage habits, thereby preliminarily improving the clustering accuracy of the clustering tree and reducing the number of unreasonable user clusters. Obtaining the feature difference degree can measure the degree of difference in user electricity usage habits between the two child nodes merged when generating a parent node, thereby determining the rationality of the parent node's clustering. Obtaining the clustering error degree can determine the difference in clustering effect when two child nodes are merged into a parent node, thereby determining the degree of reduction in clustering accuracy after the merger. Obtaining the distance difference degree can characterize the difference in clustering distance between the generation of the parent node and the generation of the corresponding two child nodes, thereby determining the rationality of the parent node's clustering based on the difference in clustering distance. Finally, the abnormal feature value of the parent node is obtained based on the feature difference degree, clustering error degree, and distance difference degree. This abnormal feature value can accurately indicate whether the electricity usage habits of users within the parent node are similar and whether the clustering result is reasonable. Pruning is performed based on the abnormal characteristic values ​​of all parent nodes in the clustering tree to obtain the final user clustering results; nodes with unreasonable and inaccurate clustering results are pruned to improve the accuracy of the final user clustering results, making the electricity consumption behavior habits of each user cluster more similar, thereby improving the accuracy of the final analysis of user behavior habits and electricity consumption process regulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 A block diagram of a concentrator intelligent fusion terminal based on user behavior analysis and regulation is provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0027] To further illustrate the technical means and effectiveness of the present invention in achieving its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effectiveness of a concentrator intelligent converged terminal based on user behavior analysis and control. In the following description, references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0028] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0029] The following describes in detail a specific solution of a concentrator intelligent fusion terminal based on user behavior analysis and regulation provided by the present invention with reference to the accompanying drawings.

[0030] See also Figure 1 , which shows a block diagram of a concentrator intelligent fusion terminal based on user behavior analysis and regulation provided by an embodiment of the present invention. The terminal includes the following modules:

[0031] The data acquisition module S1 is used to obtain different types of power consumption characteristic values ​​of users.

[0032] The concentrator collects the user's daily electricity consumption sequence, peak price electricity consumption sequence, valley price electricity consumption sequence, and daily maximum instantaneous power sequence within a collection cycle. In the embodiment of the present invention, a collection cycle is 90 days. After the electricity consumption data is collected, different types of electricity consumption characteristic values ​​of the user can be obtained based on the electricity consumption data. In the embodiment of the present invention, the electricity consumption characteristic values ​​include: daily average electricity consumption, daily electricity consumption coefficient of variation, peak-to-valley electricity ratio, and maximum load. The coefficient of variation is the ratio of daily average electricity consumption to the standard deviation of daily electricity consumption, reflecting the degree of dispersion of the user's daily average electricity consumption. The peak-to-valley electricity ratio is the user's total electricity consumption during peak prices and total electricity consumption during valley prices, reflecting the user's electricity consumption preference in different time periods. The maximum load is the average value of the daily maximum instantaneous power. All the electricity consumption characteristic values ​​obtained can characterize the user's electricity consumption behavior habits. Then, clustering can be performed based on each user's characteristic value, dividing users with different electricity consumption habits and differentially regulating the electricity consumption process. Implementers can determine the data collection type and electricity consumption characteristic values ​​based on the implementation scenario.

[0033] The power consumption feature analysis module S2 is used to obtain feature weights based on the distribution characteristics and preset importance of the same type of power consumption feature values ​​of all users; and to perform hierarchical clustering on all users based on the power consumption feature values ​​and feature weights to obtain a clustering tree.

[0034] When traditional hierarchical clustering algorithms cluster users, the difference distances obtained for different types of electricity usage characteristics are equally weighted, ultimately forming clusters based on the mean difference distances across all types of electricity usage characteristics. However, in real-world scenarios, different types of electricity usage characteristics contribute differently to user classification. If the difference between a certain type of electricity usage characteristic is greater, it means that this type of characteristic is more capable of distinguishing different users' electricity usage habits. Therefore, the weight used in calculating the difference distance should be larger, thereby improving the accuracy of user classification. If the difference between a certain type of electricity usage characteristic is smaller, it is more difficult to distinguish different users' electricity usage habits based on this type of characteristic, and the weight used in calculating the difference distance should be smaller. Therefore, an entropy weighting method can be used to assign weights to all types of electricity usage characteristics of a user, obtaining an objective weight for each type of electricity usage characteristic. Furthermore, in real-world electricity usage scenarios, some types of electricity usage characteristics, while having smaller differences, are more important in characterizing electricity usage habits and are key indicators. Calculating the difference distance based solely on objective weights may undermine the contribution of these types of electricity usage characteristics. Therefore, it is necessary to assign subjective weights to each type of electricity consumption characteristic value in combination with the actual electricity consumption behavior analysis scenario, and calculate the difference distance based on the combination of objective weights and subjective weights; therefore, the characteristic weights are obtained based on the distribution characteristics and preset importance of the same type of electricity consumption characteristic values ​​of all users.

[0035] Preferably, in an embodiment of the present invention, the step of obtaining the feature weight includes: calculating the entropy weight corresponding to each type of electricity consumption feature value of all users by an entropy weight method; it should be noted that the entropy weight method belongs to the existing technology, and the specific steps are not repeated here. The greater the amount of information of a certain type of electricity consumption feature value, the greater the entropy weight. The product of a preset first proportion value and the entropy weight of any type of electricity consumption feature value is calculated to obtain a first value; the larger the first value, the greater the objective weight of the electricity consumption feature value of that type. Calculate the product of the preset second proportion value and the preset importance corresponding to any type of electricity consumption characteristic value to obtain a second numerical value; the preset importance is determined according to the contribution of the type of electricity consumption characteristic value in analyzing the user's electricity consumption behavior habits, and the greater the contribution, the greater the preset importance; in an embodiment of the present invention, the preset importance of average daily electricity consumption is 0.4, and the average daily electricity consumption is the basic indicator, reflecting the user's electricity consumption level; the preset importance of maximum load is 0.25, and the grid capacity configuration is often based on the maximum load, which affects the planning of transmission and distribution equipment, and is more important for the evaluation of transformer load rate and line capacity; the preset importance of peak-to-valley electricity ratio is 0.2, which reflects whether the user has the ability to stagger and shift peaks, and is suitable for regulation and control demand response identification; the preset importance of the coefficient of variation of daily electricity consumption is 0.15, which can reflect the user's electricity consumption volatility and seasonal behavior, and is a non-critical factor; implementers can determine the preset importance of different types of electricity consumption characteristic values ​​according to the implementation scenario. In this embodiment of the present invention, the first and second ratios are each preset to 0.5 to adjust the ratio of objective weight to subjective weight. Implementers can determine this value based on the implementation scenario. The sum of the first and second values ​​is calculated to obtain the characteristic weight of any type of electricity usage characteristic value. A larger characteristic weight indicates a greater role for that type of electricity usage characteristic value in measuring different user electricity usage behaviors.

[0036] Furthermore, all users can be hierarchically clustered according to their electricity consumption feature values ​​and feature weights to obtain a clustering tree. Specifically, this includes: calculating the product of the difference distance of the same type of electricity consumption feature values ​​between any two users and the corresponding feature weights to obtain the weighted distance; where the difference distance of the same type is , represents the i-th electricity consumption characteristic value of the A-th user, Represents the i-th electricity consumption characteristic value of the B-th user; the larger the weighted distance, the greater the difference in electricity consumption behavior habits between the two users. Calculate the sum of all weighted distances between any two users to obtain the comprehensive distance between the two users; the larger the comprehensive distance, the greater the difference in electricity consumption behavior habits between the two users. Perform hierarchical clustering based on the comprehensive difference distances between all users to obtain a clustering tree after user clustering is completed; it should be noted that hierarchical clustering belongs to the existing technology, and the specific clustering steps will not be repeated; the difference distance is weighted to obtain the comprehensive distance and clustering is performed, so that the electricity consumption characteristic value with a greater contribution can dominate the clustering process, thereby improving the clustering accuracy in complex situations based on multiple electricity consumption characteristic values, and preliminarily reducing the possibility of users with large differences in electricity consumption behavior habits being clustered into one category.

[0037] Clustering analysis module S3 is used to obtain the degree of feature difference based on the discrete characteristics of the electricity consumption characteristic values ​​of users in any parent node of the clustering tree and the difference characteristics of the electricity consumption characteristic values ​​between users of two child nodes corresponding to any parent node; obtain the degree of clustering error of any parent node based on the difference characteristics of the clustering effects of any parent node and the corresponding two child nodes; obtain the degree of distance difference of any parent node based on the difference characteristics of the clustering distance between any parent node and the corresponding two child nodes.

[0038] After clustering is completed, it is necessary to further analyze whether there is a situation in which users with large differences in electricity consumption behavior habits are clustered into the same cluster during the user cluster merging process. In the process of clustering users' electricity consumption behavior habits, the formation of each user cluster should keep all types of electricity consumption feature values ​​relatively similar; therefore, the clustering accuracy can be analyzed by the degree of difference in user feature values ​​during the formation of user clusters. Therefore, the degree of feature difference can be obtained based on the discrete characteristics of the electricity consumption feature values ​​of users in any parent node of the clustering tree and the difference characteristics of the electricity consumption feature values ​​between the two child nodes of any parent node; preferably, in an embodiment of the present invention, the step of obtaining the degree of feature difference includes: calculating the standard deviation of the same type of electricity consumption feature values ​​of users in any parent node to obtain the degree of discreteness; the greater the degree of discreteness, the more discrete the distribution of the same type of electricity consumption feature values ​​in the arbitrary parent node, and the more obvious the volatility. The average characteristic value is obtained based on the average value of the same type of electricity consumption characteristic values ​​of users in any child node corresponding to the arbitrary parent node; the absolute value of the difference between the same type of average characteristic values ​​of the two child nodes corresponding to the arbitrary parent node is calculated; the characteristic difference is obtained; when the characteristic difference is larger, it means that the difference in electricity consumption behavior habits of users between the two child nodes is greater, then the clustering process of generating the parent node after merging will cause users with different electricity consumption behavior habits to be clustered into one category. However, since the characteristic values ​​of some types of electricity consumption fluctuate greatly, even if the characteristic difference is large, it cannot fully represent the obvious difference in the electricity consumption behavior habits of two users. Therefore, it is necessary to measure the degree of fluctuation dispersion of the electricity consumption characteristic value of this type to prevent the situation where the numerical value of the electricity consumption characteristic value of this type itself is relatively discrete and the effect of the characteristic difference is magnified. Therefore, the ratio of the characteristic difference and the degree of dispersion is calculated to obtain a third value; the third value is obtained for any type of electricity consumption characteristic value, so the sum of the third values ​​of all types is calculated and positively correlated to obtain the degree of characteristic difference of the arbitrary parent node; by calculating the ratio of the characteristic difference and the degree of dispersion, the characterization effect of the characteristic difference can be adjusted according to the fluctuation discrete characteristics of the electricity consumption characteristic value of this type itself; when the degree of characteristic difference is greater, it means that the clustering rationality of the arbitrary parent node is lower, and the more it needs to be discarded.

[0039] Furthermore, based on the difference in clustering effects between the parent node and the corresponding child nodes, when the clustering effect difference is large, it means that unreasonable clustering has occurred. Therefore, the clustering error degree of any parent node is obtained based on the difference characteristics of the clustering effects between any parent node and its corresponding two child nodes. Preferably, in an embodiment of the present invention, the step of obtaining the clustering error degree includes: calculating the average of the silhouette coefficients of the two child nodes corresponding to the arbitrary parent node; obtaining the average silhouette coefficient. It should be noted that the calculation of the silhouette coefficient belongs to the prior art and the specific steps are not repeated here. The silhouette coefficient measures the compactness within the same cluster and the separation from other clusters. The larger the silhouette coefficient, the better the compactness within the cluster and the more obvious the difference from other clusters. The difference between the silhouette coefficient of the arbitrary parent node and the average silhouette coefficient is calculated and negatively correlated to obtain the clustering error degree of the arbitrary parent node. If the silhouette coefficient of the arbitrary parent node is smaller than the average silhouette coefficient of the corresponding child node, the clustering error degree is greater, which means that the parent node's cluster compactness and separation from other clusters are worse than those of the child node, and the clustering rationality of the parent node is lower. By analyzing the difference in silhouette coefficients, we can determine whether the merged clusters have good separation and whether the cluster structure is stable. This can prevent the situation where the data of two child nodes is small, but the density within the cluster is reduced after the merger, and the boundaries with other clusters are blurred. This can improve the reliability of analyzing whether the clustering is reasonable.

[0040] Furthermore, in order to further improve the accuracy of analyzing whether the clustering structure is reasonable, the degree of distance difference of any parent node can be obtained based on the difference characteristics of the clustering distances between any parent node and the corresponding two child nodes; preferably, in an embodiment of the present invention, the step of obtaining the degree of distance difference includes: calculating the average value of the clustering distances when generating the two child nodes corresponding to any parent node to obtain the average clustering distance; the clustering distance is the distance used to determine the degree of difference between two nodes in hierarchical clustering. The larger the clustering distance, the greater the difference in the user electricity usage behavior habits between the two nodes. The difference between the clustering distance when generating any parent node and the average clustering distance is calculated and positively correlated with each other to obtain the degree of distance difference of the arbitrary parent node; when the clustering distance when generating the arbitrary parent node is greater than the average clustering distance, it means that the difference in the user electricity usage behavior habits in the two child nodes corresponding to the arbitrary parent node is greater, the greater the degree of distance difference, and the lower the rationality of the clustering result of the arbitrary parent node.

[0041] Clustering adjustment module S4 is used to obtain the abnormal feature value of any parent node according to the feature difference degree, clustering error degree and distance difference degree; prune the abnormal feature values ​​of all parent nodes in the clustering tree to obtain the final user clustering result.

[0042] After obtaining the feature difference degree, clustering error degree and distance difference degree of the parent node in the clustering tree, the clustering rationality of the parent node can be comprehensively measured; therefore, the abnormal characteristic value of any parent node is obtained according to the feature difference degree, clustering error degree and distance difference degree; preferably, in an embodiment of the present invention, the step of obtaining the abnormal characteristic value includes: calculating the average value of the feature difference degree, clustering error degree and distance difference degree to obtain the abnormal characteristic value of the arbitrary parent node; when the three are larger, the abnormal characteristic value is larger, which means that the rationality of the clustering process of obtaining the parent node by merging two child nodes is lower, and the parent node needs to be discarded more; by comprehensively measuring the clustering effect of the parent node by the three, the accuracy of measuring whether the clustering effect is reasonable is improved.

[0043] Furthermore, the cluster tree can be pruned based on the abnormal feature values ​​of all parent nodes in the cluster tree to obtain the final user clustering results. Preferably, in an embodiment of the present invention, when the abnormal feature value of any parent node in the cluster tree exceeds a preset abnormality threshold, the parent node is pruned. The user distribution result in the cluster tree after pruning is used as the final user clustering result. It should be noted that if a parent node is pruned, the parent nodes that continue to be generated when it is a child node must also be pruned and deleted. The pruned cluster tree discards unreasonable user clusters, making the final user clustering result more accurate, thereby improving the accuracy of subsequent analysis of user electricity usage habits and electricity consumption process regulation. In an embodiment of the present invention, the preset abnormality threshold is 0.7. When the preset abnormality threshold is larger, the pruning degree is smaller and the tolerance for unreasonable clusters is greater. When the preset abnormality threshold is smaller, the pruning degree is larger and the tolerance for unreasonable clusters is less. Implementers can determine the degree of pruning based on specific user classification scenarios. After obtaining the end-user clustering results corresponding to the users collected by the concentrator, different labels can be generated for different user clusters and different electricity consumption behavior habits, such as: concentrated electricity consumption during the day, abnormal load at night, peak load groups, etc. Differentiated control strategies can be formulated through different labels to accurately control the electricity consumption process. Implementers can formulate their own control strategies based on the clustering results according to the implementation scenario, which is not limited here.

[0044] In summary, the embodiments of the present invention provide a concentrator intelligent fusion terminal based on user behavior analysis and control; obtain feature weights based on the distribution characteristics and preset importance of power consumption feature values; perform hierarchical clustering based on the power consumption feature values ​​and feature weights; obtain the feature difference degree based on the discrete characteristics of the power consumption feature values ​​in any parent node of the clustering tree and the difference characteristics of the power consumption feature values ​​between users of the corresponding two child nodes; obtain the clustering error degree based on the difference characteristics of the clustering effects of any parent node and the corresponding two child nodes; obtain the distance difference degree based on the difference characteristics of the clustering distances between any parent node and the corresponding two child nodes. The present invention obtains abnormal feature values ​​based on the feature difference degree, clustering error degree, and distance difference degree; performs pruning based on the abnormal feature values ​​of the parent node to obtain the final user clustering result, thereby improving the accuracy of clustering and power consumption control.

[0045] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0046] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A concentrator intelligent fusion terminal based on user behavior analysis and control, characterized in that: The terminal includes the following modules: A data acquisition module is used to obtain different types of electricity consumption characteristic values ​​of users; The power consumption feature analysis module is used to obtain feature weights based on the distribution characteristics and preset importance of the same type of power consumption feature values ​​of all users; Performing hierarchical clustering on all users according to the electricity consumption feature values ​​and the feature weights to obtain a clustering tree; A cluster analysis module is configured to obtain a characteristic difference degree based on the discrete characteristics of the power consumption characteristic values ​​of users in any parent node of the clustering tree and the difference characteristics of the power consumption characteristic values ​​between users of two child nodes corresponding to the arbitrary parent node; obtain a clustering error degree of the arbitrary parent node based on the difference characteristics of the clustering effects of the arbitrary parent node and the corresponding two child nodes; and obtain a distance difference degree of the arbitrary parent node based on the difference characteristics of the clustering distances between the arbitrary parent node and the corresponding two child nodes. A clustering adjustment module, configured to obtain an abnormal feature value of any parent node according to the feature difference degree, the clustering error degree, and the distance difference degree; Pruning is performed according to the abnormal feature values ​​of all parent nodes in the clustering tree to obtain the final user clustering result; The step of obtaining the degree of feature difference based on the discrete features of the power consumption feature value of the user in any parent node of the clustering tree and the difference features of the power consumption feature values ​​between the users of two child nodes corresponding to the arbitrary parent node comprises: Calculate the standard deviation of the same type of electricity consumption characteristic values ​​of users in the arbitrary parent node to obtain the degree of dispersion; obtain the average characteristic value based on the average value of the same type of electricity consumption characteristic values ​​of users in any child node corresponding to the arbitrary parent node; calculate the absolute value of the difference between the same type of average characteristic values ​​of the two child nodes corresponding to the arbitrary parent node; obtain the characteristic difference value; calculate the ratio of the characteristic difference value to the degree of dispersion to obtain a third value; calculate the sum of the third values ​​of all types and positively map them to obtain the degree of characteristic difference of the arbitrary parent node.

2. The concentrator intelligent fusion terminal based on user behavior analysis and control according to claim 1 is characterized in that: The step of obtaining feature weights according to the distribution characteristics and preset importance of the same type of power consumption feature values ​​of all users includes: The entropy weight corresponding to each type of electricity consumption characteristic value of all users is calculated by the entropy weight method; the product of a preset first proportion value and the entropy weight of any type of electricity consumption characteristic value is calculated to obtain a first value; the product of a preset second proportion value and the preset importance corresponding to any type of electricity consumption characteristic value is calculated to obtain a second value; the sum of the first value and the second value is calculated to obtain the characteristic weight of the said any type of electricity consumption characteristic value.

3. The concentrator intelligent fusion terminal based on user behavior analysis and control according to claim 1, characterized in that: The step of performing hierarchical clustering on all users according to the power consumption feature values ​​and the feature weights to obtain a clustering tree includes: The weighted distance is obtained by multiplying the difference distance of the same type of electricity consumption feature values ​​between any two users by the corresponding feature weight. The combined distance between any two users is calculated to obtain the comprehensive distance between the two users. Hierarchical clustering is performed based on the comprehensive difference distances between all users to obtain a clustering tree after user clustering is completed.

4. The concentrator intelligent fusion terminal based on user behavior analysis and control according to claim 1, characterized in that: The step of obtaining the clustering error degree of the arbitrary parent node according to the difference characteristics of the clustering effects of the arbitrary parent node and the corresponding two child nodes includes: Calculate the average value of the silhouette coefficients of the two child nodes corresponding to the arbitrary parent node; obtain the average silhouette coefficient; calculate the difference between the silhouette coefficient of the arbitrary parent node and the average silhouette coefficient and perform negative correlation mapping to obtain the clustering error degree of the arbitrary parent node.

5. The concentrator intelligent fusion terminal based on user behavior analysis and control according to claim 1, characterized in that: The step of obtaining the distance difference degree of the arbitrary parent node according to the difference characteristics of the cluster distances between the arbitrary parent node and the corresponding two child nodes includes: For the two child nodes corresponding to the arbitrary parent node, the average value of the clustering distances when the two child nodes are generated is calculated to obtain the average clustering distance; the difference between the clustering distance when the arbitrary parent node is generated and the average clustering distance is calculated and positively correlated with each other to obtain the distance difference degree of the arbitrary parent node.

6. The concentrator intelligent fusion terminal based on user behavior analysis and control according to claim 1, characterized in that: The step of obtaining the abnormal feature value of any parent node according to the feature difference degree, the clustering error degree and the distance difference degree comprises: The average values ​​of the feature difference degree, the clustering error degree, and the distance difference degree are calculated to obtain the abnormal feature value of the arbitrary parent node.

7. The concentrator intelligent fusion terminal based on user behavior analysis and control according to claim 1, characterized in that: The step of pruning according to the abnormal feature values ​​of all parent nodes in the clustering tree to obtain the final user clustering result includes: When the abnormal feature value of any parent node in the clustering tree exceeds a preset abnormal threshold, the arbitrary parent node is pruned; and the user distribution result in the clustering tree after pruning is used as the final user clustering result.

8. The concentrator intelligent fusion terminal based on user behavior analysis and control according to claim 1, characterized in that: The power consumption characteristic values ​​include: average daily power consumption, coefficient of variation of daily power consumption, peak-to-valley power consumption ratio, and maximum load.

Images