Knowledge-driven unsupervised user clustering model accuracy rate comprehensive evaluation method in electric power marketing field
By proposing an unsupervised user clustering model and its evaluation method in the field of power marketing, and conducting multi-dimensional evaluation with domain knowledge, the problem of accuracy evaluation of unsupervised clustering models in the power industry is solved, and the accurate classification of customers is achieved and Efficient evaluation of the model.
Patent Information
- Application Number
- CN202411473390.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2025-02-18
AI Technical Summary
In the power industry, how to accurately evaluate the accuracy of the unsupervised learning clustering model, especially in the environment where user data is large and lacks clear labels, has become an urgent problem. Existing evaluation methods rely on external verification sets or manual annotations, are difficult to implement and expensive.
A knowledge-driven unsupervised user clustering model method in the field of power marketing and its comprehensive evaluation method for accuracy is proposed. By collecting user data, unsupervised learning algorithms are used for clustering analysis, and combining domain knowledge, multi-dimensional evaluation standards are designed, including electricity consumption stability, electricity bill sensitivity, payment behavior pattern and activity, etc., to perform clustering models. evaluating accuracy.
It realizes accurate classification of customers in the power marketing field and multi-dimensional accurate evaluation of classification models. It does not require external verification sets, and uses the statistical characteristics and business logic consistency of the clustering results to evaluate, improving the accuracy and practicality of the clustering model. sex.
Smart Images
Figure CN119477402A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing and machine learning, and in particular to unsupervised learning cluster analysis of user data in the power industry. Background Art
[0002] In the power industry, user clustering analysis is an important means to formulate precise marketing strategies, optimize service quality and improve operational efficiency. However, how to accurately evaluate the accuracy of unsupervised learning clustering models, especially in the power marketing data environment with huge amounts of user data and lack of clear labels, has become an urgent problem to be solved. Existing evaluation methods often rely on external validation sets or manual labeling, which is difficult and costly to implement in large-scale, high-dimensional user data in the power industry. Moreover, traditional clustering models often ignore the guiding role of domain knowledge when evaluating accuracy, resulting in inaccurate evaluation results or difficulty in direct application to actual business scenarios. Summary of the invention
[0003] The present invention proposes a knowledge-driven unsupervised user clustering model method in the field of power marketing and a comprehensive accuracy evaluation method thereof. The specific technical solution is as follows:
[0004] First, the knowledge-driven unsupervised user clustering model method in the power marketing field includes the following processes:
[0005] S01: Collect user data of a certain area in the power industry over a period of time. The specific data of a single user includes: average monthly electricity consumption, average monthly electricity fee, average number of days between each payment time in the year, total number of payments in the year, payment channel with the most payments, average monthly payment amount, most commonly used payment channel, average monthly login times of users, average monthly login duration of users, average monthly login days of users, historical activities participated by users, and number of times participated;
[0006] The average monthly electricity consumption can be obtained by dividing the total annual electricity consumption by 12; the average monthly electricity fee can be obtained by dividing the annual electricity fee by 12; the average monthly payment can be obtained by dividing the annual payment by 12; the average monthly login times of users can be obtained by dividing the annual login times of users by 12; the average monthly login days of users can be obtained by dividing the annual login days of users by 12;
[0007] S02: Use an unsupervised learning algorithm to perform cluster analysis on the preprocessed data to generate user clustering results.
[0008] Preferably, the following process is included between S01 and S02:
[0009] S01-02: Data preprocessing, including data cleaning, denoising, missing value processing and standardization, to ensure data quality.
[0010] Preferably, the unsupervised learning algorithm is a Gaussian mixture clustering model.
[0011] Secondly, the comprehensive evaluation method of the accuracy of the above clustering model includes the following process:
[0012] S03: Calculate the standard deviation or coefficient of variation of the average monthly total electricity consumption of all users in the same cluster, and evaluate the differences in electricity consumption stability among different user groups in the clustering results; the calculation formula is as follows:
[0013]
[0014] Among them, X i is the average monthly electricity consumption, is the average monthly electricity consumption of users in the cluster, and n is the number of users in the cluster.
[0015] The smaller the standard deviation or the lower the coefficient of variation, the more stable the electricity consumption of the users in the cluster. It is used to evaluate the differences in electricity consumption stability among different user groups in the clustering results.
[0016] Thirdly, the comprehensive evaluation method of the accuracy of the above clustering model includes the following process:
[0017] S04: Calculate the correlation coefficient between the average monthly electricity bill and the average monthly electricity consumption of all users in the same cluster to evaluate the user's sensitivity to changes in electricity bills;
[0018] The correlation coefficient calculation formula is as follows:
[0019]
[0020] Among them, X i and Y i are the average monthly electricity consumption and the average monthly electricity bill, and are the average monthly electricity consumption and monthly electricity charges in the cluster, respectively, and n is the number of users in the cluster.
[0021] Fourthly, the comprehensive evaluation method of the accuracy of the above clustering model includes the following process:
[0022] S05: Calculate the average number of days between each payment time and the average number of total payment times in the year for all users in the same cluster; count the payment channels with the most payment times; and evaluate the payment behavior patterns of users;
[0023]
[0024] Where D i is the number of days between two consecutive payments of user i, and n is the number of users in the cluster
[0025] S06: Calculate the comprehensive activity index of all users in the same cluster and evaluate the differences in user activity among different user groups in the clustering results;
[0026] The activity index calculation formula is as follows:
[0027]
[0028] Among them, α, β, and γ are weight coefficients, which are adjusted according to the importance of the business; M is the total number of activities;
[0029] Calculate the standard deviation or coefficient of variation of the activity index to determine whether it is less than the preset threshold; if it is smaller, it means that the degree of dispersion of users in the cluster is small and the consistency is high;
[0030] S07: Draw a distribution graph of the user activity index within the cluster, and observe and analyze the following forms and characteristics of the graph: observe whether the activity indexes are concentrated in a certain range. If so, it means that the consistency of users within the cluster is high; observe whether the distribution graph presents obvious normal distribution, skewed distribution or bimodal distribution forms. If the distribution form is relatively regular, such as normal distribution, it means that the consistency of users within the cluster is high.
[0031] Fifthly, the comprehensive evaluation method of the accuracy of the above clustering model includes the following process:
[0032] S08: Use the silhouette coefficient to calculate the similarity of the electricity consumption behavior, payment behavior, activity, etc. of users in the same cluster to ensure the internal consistency of the clustering results; the electricity consumption behavior and payment behavior refer to the average monthly electricity consumption, the average monthly electricity fee, the average number of payment interval days, the total number of payments, the most common payment channel, and the average monthly payment amount; the silhouette coefficient calculation method is as follows:
[0033] For each sample point, calculate the average distance a from other sample points in the same cluster, which represents the cohesion; calculate the average distance b from all points in the nearest cluster, which represents the separation; then calculate the silhouette coefficient s, the formula is as follows:
[0034] s=(ba) / max(a,b);
[0035] The silhouette coefficient of the entire cluster is the average of the silhouette coefficients of all sample points; the silhouette coefficient value range is between -1 and 1. A value close to 1 indicates that the sample clustering is reasonable, the intra-cluster distance is small and the inter-cluster distance is large; a value close to 0 indicates that the sample clusters overlap; a value close to -1 indicates that the sample is incorrectly assigned to an adjacent cluster.
[0036] In the sixth aspect, the comprehensive evaluation method of the accuracy of the above clustering model includes the following process:
[0037] Analyze the differences between different clusters in the above evaluation criteria to ensure the distinction between clustering results; use the Davies-Bouldin index DBI to evaluate the differences between clusters; specifically, evaluate the clustering effect by calculating the maximum value of the ratio of the separation between each two clusters to the compactness within the cluster; the smaller the DBI value, the better the clustering effect.
[0038] The calculation formula of DBI is as follows:
[0039]
[0040] Where N is the number of clusters, S i is the Euclidean distance from the element in the i-th class to the center of the class, / / w i -w j / / 2 is the Euclidean distance between the centers of the i-th and j-th clusters.
[0041] Dynamic feedback and optimization: Based on the results of internal validation and evaluation, the accuracy and shortcomings of the clustering model are analyzed qualitatively and quantitatively.
[0042] Quantitative analysis: Quantitative analysis is performed on the evaluation criteria of clustering results, and statistics such as the mean, standard deviation, and median of each indicator are calculated to understand the overall distribution and dispersion of the clustering results.
[0043] Qualitative analysis: Combine business logic and actual conditions to conduct qualitative analysis of clustering results. Identify the rationality and irrationality of clustering results and explore the reasons behind them.
[0044] If some clustering results are found to be inaccurate or overlapped, adjust the parameters of the clustering algorithm, such as increasing the number of clusters K, adjusting the neighborhood radius ε, etc., or consider introducing new features or adjusting the weights of existing features to improve the clustering effect. After implementing the optimization measures, re-conduct clustering analysis and evaluation to form a closed-loop optimization process to continuously improve the accuracy and practicality of the clustering model.
[0045] Effects of the present invention relative to the prior art: The present invention combines the data generated in the field of power marketing with a clustering evaluation model to achieve accurate classification of customers in the field of power marketing and accurate multi-dimensional evaluation of the classification model. In the embodiment, an internal verification method that does not require an external verification set is designed, and the accuracy and stability of the clustering model are evaluated by using the statistical characteristics and business logic consistency of the clustering results themselves. In the embodiment, a dynamic feedback mechanism based on the evaluation results is established, and through parameter adjustment and optimization strategies, continuous iterative optimization of the clustering model is achieved to improve its practicality and adaptability in the field of power marketing. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 The present invention is a flowchart of a user clustering model classification method and an evaluation method thereof in the field of power marketing in an embodiment of the present invention. DETAILED DESCRIPTION
[0047] The Gaussian mixture clustering model was used to perform cluster analysis on user data in the power industry of a certain province. The data was collected from a certain region in the past year, covering multiple dimensions such as the user's average monthly electricity consumption, average monthly electricity bill, payment behavior, and login behavior. After data preprocessing, including data cleaning, denoising, missing value processing, and standardization, the data quality was ensured. Finally, the user data was divided into three categories: A, B, and C.
[0048] In this embodiment,
[0049] Among them, α, β, and γ are weight coefficients, which are 0.4, 0.4, and 0.2 respectively.
[0050] The specific data are as follows:
[0051] Category A user data:
[0052] Standard deviation of monthly average electricity consumption: 15kWh;
[0053] Coefficient of variation of average monthly electricity consumption: 0.08;
[0054] Correlation coefficient between average monthly electricity bill and average monthly electricity consumption: 0.85;
[0055] Average number of days between payments: 28;
[0056] Activity index standard deviation: 0.5;
[0057] User characteristics: The electricity consumption is very stable with minimal fluctuations; they are sensitive to changes in electricity charges; their payment behavior is very regular, with a high on-time payment rate of 98%; their activity level is very high;
[0058] User types may be described as: highly stable residential or commercial users, with regular electricity usage behavior, timely payments, and high activity.
[0059] Category B user data:
[0060] Standard deviation of monthly average electricity consumption: 35kWh;
[0061] Coefficient of variation of average monthly electricity consumption: 0.15;
[0062] Correlation coefficient between average monthly electricity bill and average monthly electricity consumption: 0.92;
[0063] Average number of days between payments: 42;
[0064] Activity index standard deviation: 1.2;
[0065] User characteristics: Electricity consumption fluctuates the most, and may include a large number of high-energy consumption or low-energy consumption users, and electricity consumption behavior is unstable; sensitivity to changes in electricity charges is relatively low; payment behavior is the most irregular, and the proportion of on-time payments is low; activity is relatively low.
[0066] User type may be described as: moderately volatile commercial or industrial user, sensitive to electricity costs, relatively regular payment behavior, and moderately active.
[0067] Category C user data:
[0068] Standard deviation of monthly average electricity consumption: 80kWh;
[0069] Average monthly electricity consumption coefficient of variation: 0.3;
[0070] Correlation coefficient between average monthly electricity bill and average monthly electricity consumption: 0.75;
[0071] Average number of days between payments: 60;
[0072] Activity index standard deviation: 2;
[0073] User characteristics description: electricity consumption is relatively large, but still acceptable; very sensitive to changes in electricity charges; payment behavior is relatively irregular, but the proportion of on-time payments remains at 90%; there are certain differences in activity levels.
[0074] User types may be described as: low stability or emerging users, unstable electricity usage behavior, low sensitivity to electricity charges, irregular payments, and low activity.
[0075] The clustering model evaluation process is as follows:
[0076] The first aspect: electricity consumption stability assessment
[0077] Class A users: The standard deviation of the average monthly electricity consumption is 15 kWh. This shows that the electricity consumption of Class A users is very stable with minimal fluctuations, which meets the rules for high-power stable users.
[0078] Class B users: The standard deviation of the average monthly electricity consumption is 35 kWh. The electricity consumption of Class B users fluctuates relatively greatly, but is still within an acceptable range.
[0079] Class C users: The standard deviation of the average monthly electricity consumption is 80 kWh. Class C users have the largest fluctuation in electricity consumption, and may include a large number of high-energy or low-energy users, and their electricity consumption behavior is unstable.
[0080] Second aspect: Electricity price sensitivity assessment
[0081] Class A users: The correlation coefficient between the average monthly electricity bill and the average monthly electricity consumption is 0.85, indicating that Class A users are sensitive to changes in electricity bills, but not as high as other types of users. This may be related to the stability of their electricity consumption.
[0082] Class B users: The correlation coefficient between the average monthly electricity bill and the average monthly electricity consumption is 0.92, indicating that Class B users are very sensitive to changes in electricity bills, and the electricity bill is highly positively correlated with electricity consumption, which meets the rules for electricity bill-sensitive users.
[0083] Class C users: The correlation coefficient between the average monthly electricity bill and the average monthly electricity consumption is 0.75, indicating that Class C users are relatively less sensitive to changes in electricity bills, but there is still a certain correlation.
[0084] The third aspect: payment behavior pattern assessment
[0085] Class A users: The average payment interval is 28 days, and the total number of payments is 13 times a year. The payment channels are mainly online banking and third-party payment platforms. Class A users have very regular payment behavior and tend to use online payment channels. The proportion of on-time payment is as high as 98%, which meets the rules of timely payment.
[0086] Class B users: The average payment interval is 42 days, with a total of 9 payments per year. The payment channels are relatively diverse, including online and offline payments. Class B users' payment behavior is relatively irregular, but the proportion of on-time payments remains above 90%.
[0087] Class C users: The average payment interval is 60 days, and the total number of payments is 6 times a year. The payment channels are mainly bank counters and self-service payment terminals. Class C users have the most irregular payment behavior, and the proportion of on-time payment is low, only 75%.
[0088] The fourth aspect: activity assessment
[0089] Class A users: The activity index is relatively high and the dispersion is relatively small, indicating that the activity of Class A users is very high and consistent, which meets the active user rules.
[0090] Category B users: The activity index is at a medium level with a moderate degree of dispersion, indicating that there are certain differences in the activity of Category B users, but overall it is still relatively stable.
[0091] Category C users: The activity index is low and has a large dispersion, indicating that Category C users have low activity and poor consistency.
[0092] Aspect 5: Silhouette coefficient evaluation
[0093] Class A users: The silhouette coefficient is 0.88, which indicates that the clustering result of class A users is very reasonable, with small intra-cluster distance and large inter-cluster distance.
[0094] Class B users: The silhouette coefficient is 0.78, indicating that the clustering result of class B users is relatively reasonable, but there is a certain overlap.
[0095] Class C users: The silhouette coefficient is 0.65, indicating that the clustering result of class C users is relatively poor, and there may be some users who are misassigned.
[0096] Aspect 6: Davies-Bouldin Index (DBI) Assessment
[0097] The overall DBI value is 0.42, indicating that the clustering effect is good overall. Among them, the separation between category A users and category B users is high, and the intra-cluster compactness is large; while the separation between category C users and the other two categories of users is relatively low, but still within an acceptable range.
[0098] Seventh aspect: Verification of knowledge in the field of integrated power marketing business
[0099] Rules for stable users with high electricity consumption: The proportion of Class A users who meet the conditions for stable users with high electricity consumption is 90%, indicating that the classification of Class A as a stable user with high electricity consumption is very accurate.
[0100] Electricity-rate-sensitive user rules: The proportion of Class B users that meet the definition of electricity-rate-sensitive users is 85%, indicating that the Class B clustering results have a strong ability to identify electricity-rate-sensitive users.
[0101] Timely payment rule: The on-time payment rate for Class A users is as high as 98%, for Class B users it is 90%, and for Class C users it is 75%, which is in line with the expectations of the timely payment rule.
[0102] Payment channel diversity rules: Category A users mainly use online banking and third-party payment platforms, Category B users have more diverse payment channels, and Category C users mainly use bank counters and self-service payment terminals, which is in line with the expectations of the payment channel diversity rules.
[0103] Large-amount payment and prepayment behavior rules: A certain proportion of Class A users make large-amount payments and tend to pay their electricity bills in advance; Class B and Class C users have relatively few behaviors in this regard.
[0104] Active user rules: The proportion of active users among Class A users is relatively high, Class B users are at a medium level, and Class C users are relatively low, which is in line with the expectations of the active user rules.
[0105] Through quantitative and qualitative analysis, it is found that the clustering results of Class C users are relatively poor and there are overlaps. Therefore, it is considered to increase the number of clusters K or adjust parameters such as the neighborhood radius ε to improve the clustering effect. At the same time, it is planned to introduce new features, such as the user's historical arrears records, peak and valley periods of electricity consumption, etc., to enrich the input information of the clustering model and improve the accuracy and practicality of the clustering results. After the optimization measures are implemented, clustering analysis and evaluation will be re-performed to form a closed-loop optimization process to continuously improve the accuracy and practicality of the clustering model.
[0106] In summary, through the cluster analysis and evaluation of the user data of the power industry in a province, the users were successfully divided into three categories: A, B, and C, and the clustering results were evaluated in detail through multiple dimensions. The evaluation results show that Class A users have higher power stability, electricity fee sensitivity, and activity; Class B users have a medium performance in all aspects; Class C users have poor performance in all aspects. The evaluation results are of great significance for power companies to formulate targeted marketing strategies and improve user satisfaction.
[0107] In the present invention, the knowledge verification of the integrated power marketing business domain includes the following aspects:
[0108] Rules for stable users with high electricity consumption: Users in the stable users cluster should show high electricity consumption stability and low sensitivity to electricity bills. The proportion of users who meet the above conditions to the total users of the cluster is calculated. The higher the proportion, the more accurate the classification of the cluster as a stable user with high electricity consumption.
[0109] Electricity-rate-sensitive users rule: Electricity-rate-sensitive users should show a high correlation between electricity charges and electricity consumption, with a correlation coefficient close to 1 or -1, and may be accompanied by a high proportion of electricity charges, with a high ratio of "total annual payment / 12" to "total annual electricity consumption / 12". Count the proportion of users who meet the definition of electricity-rate-sensitive users and evaluate the ability of clustering results to identify such users.
[0110] Timely payment rule: By analyzing the time difference between the user's "payment time" and the "due date" of the electricity bill, it is determined whether the user pays the electricity bill on time to avoid late payment fees or overdue fees. Calculate the proportion of users in the cluster who pay on time, that is, complete payment before the due date of the electricity bill. The higher the proportion, the better the performance of users in the cluster in terms of timely payment.
[0111] Payment channel diversity rule: Analyze whether users use multiple channels to pay electricity bills, including online payment, such as online banking, third-party payment platforms, offline payment, such as bank counters, self-service payment terminals, and third-party deductions, etc. Count the proportion and frequency of users using different payment channels to evaluate users' preference and diversity of payment channels.
[0112] Large-amount payment and prepayment behavior rules: Pay attention to whether users have large-amount payment behavior and whether they tend to use prepayment to pay electricity bills. Large-amount payments may be related to users' special needs or changes in economic conditions, while prepayment reflects users' payment habits and initiative in electricity bill management. Set thresholds for large-amount payments, count the proportion and frequency of large-amount payments by users; at the same time, analyze the proportion of users who use prepayment and compare it with postpayment.
[0113] Active user rule: Active users should have a higher activity index. The proportion of active users is calculated and the recognition effect of clustering results on user activity is evaluated.
[0114] Internal verification is performed using the statistical properties of clustering results—intra-cluster similarity, inter-cluster difference, and business logic consistency: Statistical property verification ensures that the clustering results are mathematically and statistically meaningful, that is, the clustering is reasonable and stable. This helps to eliminate accidental clustering results caused by algorithm randomness. Business logic consistency verification combines clustering results with actual business scenarios to verify their practical application value. This requires the evaluator to have deep business knowledge and a deep understanding of the clustering results.
Claims
1. An unsupervised user clustering model method driven by knowledge in the field of power marketing, characterized by: The process includes the following: S01: Collect user data of a certain area in the power industry over a period of time. The specific data of a single user includes: average monthly electricity consumption, average monthly electricity fee, average number of days between each payment time in the year, total number of payments in the year, payment channel with the most payments, average monthly payment amount, most commonly used payment channel, average monthly login times of users, average monthly login duration of users, average monthly login days of users, historical activities participated by users, and number of times participated; S02: Use an unsupervised learning algorithm to perform cluster analysis on the preprocessed data to generate user clustering results.
2. According to claim 1, the user clustering model classification method in the field of power marketing is characterized by: The following process is included between S01 and S02: S01-02: Data preprocessing, including data cleaning, denoising, missing value processing and standardization, to ensure data quality.
3. The user clustering model classification method in the power marketing field according to claim 1 or 2 is characterized in that: The unsupervised learning algorithm is a Gaussian mixture clustering model.
4. A comprehensive evaluation method for the accuracy of an unsupervised user clustering model driven by knowledge in the field of power marketing for evaluating the user clustering model method of claim 1, characterized in that: The process includes the following: S03: Calculate the standard deviation or coefficient of variation of the average monthly total electricity consumption of all users in the same cluster, and evaluate the differences in electricity consumption stability among different user groups in the clustering results; the calculation formula is as follows: Among them, X i is the average monthly electricity consumption, is the average monthly electricity consumption of users in the cluster, and n is the number of users in the cluster.
5. A comprehensive evaluation method for the accuracy of an unsupervised user clustering model driven by knowledge in the field of power marketing for evaluating the user clustering model method of claim 1, characterized in that: The process includes the following: S04: Calculate the correlation coefficient between the average monthly electricity bill and the average monthly electricity consumption of all users in the same cluster to evaluate the user's sensitivity to changes in electricity bills; The correlation coefficient calculation formula is as follows: Among them, X i and Y i are the average monthly electricity consumption and the average monthly electricity bill, and are the average monthly electricity consumption and monthly electricity charges in the cluster, respectively, and n is the number of users in the cluster.
6. A comprehensive evaluation method for the accuracy of an unsupervised user clustering model driven by knowledge in the field of power marketing for evaluating the user clustering model method of claim 1, characterized in that: The process includes the following: S05: Calculate the average number of days between each payment time and the average number of total payment times in the year for all users in the same cluster; count the payment channels with the most payment times; and evaluate the payment behavior patterns of users; Where D i is the number of days between two consecutive payments of user i, and n is the number of users in the cluster; S06: Calculate the comprehensive activity index of all users in the same cluster and evaluate the differences in user activity among different user groups in the clustering results; The activity index calculation formula is as follows: Among them, α, β, and γ are weight coefficients, and M is the total number of activities; Calculate the standard deviation or coefficient of variation of the activity index to determine whether it is less than the preset threshold; if it is smaller, it means that the degree of dispersion of users in the cluster is small and the consistency is high; S07: Draw a distribution graph of the user activity index within the cluster, and observe and analyze the following forms and characteristics of the graph: observe whether the activity indexes are concentrated in a certain range. If so, it means that the consistency of users within the cluster is high; observe whether the distribution graph presents an obvious normal distribution, skewed distribution or bimodal distribution. If the distribution pattern is relatively regular, it means that the consistency of users within the cluster is high.
7. A comprehensive evaluation method for the accuracy of an unsupervised user clustering model driven by knowledge in the power marketing field for evaluating the user clustering model method of claim 1, characterized in that: The process includes the following: S08: Use the silhouette coefficient to calculate the similarity of electricity consumption behavior, payment behavior, and activity of users in the same cluster to ensure the internal consistency of the clustering results; the silhouette coefficient calculation method is as follows: For each sample point, calculate its average distance a to other sample points in the same cluster and its average distance b to all points in the nearest cluster, and then calculate the silhouette coefficient s. s=(ba) / max(a,b); The silhouette coefficient of the entire cluster is the average of the silhouette coefficients of all sample points; the silhouette coefficient value range is between -1 and 1. A value close to 1 indicates that the sample clustering is reasonable, the distance within the cluster is small, and the distance between clusters is large; a value close to 0 indicates that the sample clusters overlap; Values close to -1 indicate that the sample was incorrectly assigned to an adjacent cluster.
8. A comprehensive evaluation method for the accuracy of an unsupervised user clustering model driven by knowledge in the field of power marketing for evaluating the user clustering model method of claim 1, characterized in that: The Davies-Bouldin index (DBI) was used to evaluate the differences between clusters. The smaller the DBI value, the better the clustering effect. The specific process includes the following: The maximum value DBI of the ratio of the separation between each two clusters to the compactness within the cluster is calculated using the following formula: Where N is the number of clusters, S i is the Euclidean distance from the element in the i-th class to the center of the class, / / w i -w j / / 2 is the Euclidean distance between the centers of the i-th and j-th classes.