High-value power customer-oriented service optimization method and equipment

Abstract

The invention relates to the technical field of information, and provides a high-value power customer oriented service optimization method and equipment, and the method comprises the steps: judging a power customer feature correlation coefficient through a correlation coefficient matrix, and removing redundant features; marking high-value power customers according to the power customer feature set after dimension reduction; obtaining dynamic behavior data of the high-value power customers from the high-value customer group to obtain dynamic feature vectors of behaviors of the high-value power customers; aiming at the dynamic feature vector, the long-short-term memory network model outputs a behavior prediction result; complaint records and interaction logs of high-value power customers are obtained, and complex appeals are marked; determining a priority service object by adopting a weighted scoring method; and adopting a recommendation algorithm to generate a personalized service strategy for the priority service object. According to the method, accurate identification, behavior prediction and service optimization of high-value power customers are realized, the service quality and satisfaction of the power customers are effectively improved, and technical support is provided for fine operation and differentiated services of power enterprises.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a service optimization method and device for high-value power customers. Background Art

[0002] The analysis of power supply enterprise customer demands is a core research area for optimizing service quality and enhancing customer satisfaction in the power industry. Its importance lies in its direct relation to the operation efficiency and market competitiveness of the enterprise. With the advancement of power market reform, customer demands have become increasingly diverse, and demand analysis has become a key link in promoting precise services and resource allocation. However, many current solutions still remain at the qualitative analysis level, relying on manual experience or simple statistics, and it is difficult to comprehensively capture the complexity and dynamic changes of customer behavior. This results in enterprises often being unable to accurately identify high-value customers or quickly respond to their core demands when faced with a large amount of customer data, and the service efficiency and customer experience are significantly restricted. The limitations of existing methods are mainly reflected in insufficient data processing depth and single analysis dimension. Traditional methods mostly classify customers using a single indicator such as electricity consumption or the number of complaints, ignoring the multi-dimensional characteristics and potential correlations of customer behavior. This extensive analysis is difficult to adapt to the personalized trend of customer demands and cannot provide a scientific decision-making basis for enterprises. In this field, the core challenges focus on how to effectively integrate multi-dimensional customer data and transform it into actionable quantitative indicators. Specifically, the heterogeneity of customer characteristics, the dynamics of behavior, and the complexity of demands are the three major technical factors. Since customer characteristics cover multiple aspects such as electricity consumption habits and payment records, the data sources are scattered and the formats are inconsistent, making integration difficult; customer behavior changes over time, and static models are difficult to reflect their real needs; the complexity of demands is difficult to simply classify due to multiple factors such as emotions and scenarios involved. These unresolved technical factors lead to unique problems for enterprises in identifying high-value customers and optimizing service strategies, and there is an urgent need for systematic quantitative methods. Therefore, how to numerically process customer characteristics, behaviors, and demands in multiple dimensions and construct a scientific evaluation system to accurately identify high-value customers and improve their satisfaction has become a key issue in the field of power supply enterprise customer demand analysis. Summary of the Invention

[0003] To achieve the object of the present invention, in a first aspect, the present invention provides a service optimization method for high-value power customers, mainly including: Multidimensional data is obtained from the electricity consumption records, payment data, and complaint logs of electricity customers, and the multidimensional data is preprocessed to obtain an electricity customer dataset with a unified data structure. For the electricity customer dataset with a unified data structure, the corr function of the pandas library is used to calculate the correlation coefficients between the features of electricity customers, and a Pearson correlation coefficient matrix is constructed. Based on the Pearson correlation coefficient matrix, the principal component analysis method is used to generate the principal component feature vectors of electricity customers, and the electricity customer feature set after dimensionality reduction is obtained. According to the electricity customer feature set after dimensionality reduction, the Euclidean norm of the principal component feature vectors of electricity customers is calculated to obtain the eigenvalues of electricity customers, and it is judged whether the eigenvalues exceed a preset threshold T1. If so, the electricity customer is determined to be a high-value electricity customer. The electricity consumption and complaint frequency of the high-value electricity customers are obtained from the electricity customer dataset, and the time series analysis technology is used to calculate the change rates of the electricity consumption and complaint frequency within a continuous time window to obtain the dynamic feature vectors of the high-value electricity customers. For the dynamic feature vectors of the high-value electricity customers, a multidimensional feature matrix of the high-value electricity customers is constructed according to the time series, and the multidimensional feature matrix is input into a pre-trained long short-term memory network model to output the behavior prediction results of the high-value electricity customers. The complaint records and interaction logs of the high-value electricity customers are obtained from the electricity customer dataset, and the complaint records and interaction logs of the high-value electricity customers are input into a preset sentiment analysis model for sentiment analysis to obtain the sentiment analysis results of the high-value electricity customers. An emotion analysis feature vector is constructed through the emotion analysis results and the behavior prediction results, and the emotion analysis feature vector is input into a pre-constructed decision tree model for appeal complexity classification to obtain the appeal complexity classification results. A weighted scoring method is used to comprehensively score the electricity consumption, behavior dynamics, and appeal complexity classification results of high-value electricity customers, and the service priority classification results of high-value electricity customers are determined based on the comprehensive score. If the service priority classification result of the high-value electricity customer is a priority service object, a recommendation algorithm is used to generate a personalized service strategy for the high-value electricity customer. If the estimated response time of the personalized service strategy is less than the preset threshold T3, it is determined as an executable strategy, and the final service optimization plan is obtained.

[0004] In a second aspect, the present invention also provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the method described in any one of the above are implemented.

[0005] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: The present invention discloses a service optimization method and device for high-value power customers. The method integrates customer electricity consumption, payment, and complaint data, and uses the principal component analysis method and clustering algorithm to identify high-value customers. Further, time series analysis and long short-term memory network model are used to predict customer behavior changes, and the long short-term memory network model is combined to analyze the complexity of customer demands. Finally, a quantitative evaluation system is constructed based on the weighted scoring method to generate personalized service strategies for priority service objects. The present invention realizes the accurate identification, behavior prediction, and service optimization of high-value customers, effectively improves the customer service quality and satisfaction, and provides technical support for the refined operation and differentiated service of power enterprises. Brief Description of the Drawings

[0006] Figure 1 It is a flowchart of the service optimization method for high-value power customers of the present invention.

[0007] Figure 2 It is a structural schematic block diagram of the computer device of the embodiment of the present invention. Detailed Embodiment

[0008] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0009] As Figure 1 , the service optimization method for high-value power customers in this embodiment may specifically include: S101. Obtain multi-dimensional data from the electricity consumption records, payment data, and complaint logs of power customers, and preprocess the multi-dimensional data to obtain a power customer data set with a unified data structure.

[0010] Retrieve multi-dimensional data from power customer electricity consumption records, payment data, and complaint logs. Use the Pandas library to separate structured and unstructured fields in the multi-dimensional data to obtain an initial dataset. Process the initial dataset using the Pandas library. If missing values are detected, fill numerical fields using the mean filling method and categorical fields using the mode filling method to obtain a filled dataset. For the filled dataset, if the values of numerical fields are detected to exceed the preset numerical threshold range, remove the outliers outside the numerical threshold range to obtain a cleaned dataset. Use StandardScaler to standardize the cleaned dataset and convert fields in different formats to a preset structure using a unified encoding rule to obtain a standardized dataset. Extract features including customer identification, electricity consumption, payment amount, and complaint frequency from the standardized dataset, and use the groupby function of the Pandas library to generate multi-dimensional feature vectors to obtain an aggregated dataset. Use the K-means clustering algorithm to classify the aggregated dataset, divide customer groups according to power customer electricity consumption behavior and complaint behavior to obtain a classified power customer dataset. Use the Apriori algorithm to analyze the classified power customer dataset to determine the correlation between electricity consumption patterns and complaint patterns among power customer groups to obtain a power customer dataset with a unified data structure.

[0011] Exemplarily, in the power customer electricity consumption business scenario, the processing and analysis of multi-dimensional data can effectively mine the power customer behavior patterns and improve the service quality. Exemplarily, multi-dimensional data is obtained from the power customer electricity consumption records, payment data, and complaint logs. The power customer electricity consumption records contain structured fields such as power customer identification and monthly electricity consumption, as well as unstructured fields such as remarks; the payment data includes payment amount and payment time; the complaint logs include complaint content and time. The Pandas library is used to separate the fields. Structured fields such as electricity consumption and payment amount are directly extracted, and unstructured fields such as complaint text are converted into category labels through keyword extraction. For example, the record of power customer A contains electricity consumption of 500 kWh, payment amount of 600 yuan, and complaint content "unstable voltage". After separation, the initial data set is obtained, which contains structured fields of electricity consumption, payment amount, and the complaint category "voltage problem" converted from unstructured data. In a possible implementation, the missing values in the initial data set are processed. For numerical fields such as electricity consumption, if missing, the mean value is used for filling. For example, if the electricity consumption of power customer A is missing in a certain month, the mean value of 400 kWh of other months of power customer A is calculated for filling. For categorical fields such as complaint category, if missing, the mode is used for filling. For example, if "no complaint" appears the most, it is filled with this value. After filling, the data set is complete, which is conducive to subsequent analysis and ensures data consistency. Specifically, when cleaning the data set, the outliers in the numerical fields are detected. The preset threshold range of electricity consumption is 50 to 1000 kWh. If the electricity consumption of power customer A is 2000 kWh, it is determined as abnormal and excluded. After cleaning, the data set is more in line with the actual business scenario and improves the data reliability. Preferably, for standardization processing, the StandardScaler is used to convert numerical fields such as electricity consumption and payment amount into a standard distribution with a mean of 0 and a variance of 1. For categorical fields such as payment status, they are uniformly encoded as numerical values. For example, "paid" is encoded as 1 and "unpaid" is encoded as 0. The standardized data set has a unified format, which is convenient for subsequent algorithm processing. For example, after standardization, fields with different dimensions can be directly compared, which improves the clustering effect and unifies the dimensions and formats, facilitating algorithm processing. Features are extracted and multi-dimensional feature vectors are generated. The groupby function of Pandas is used to aggregate by power customer identification, and the average electricity consumption, total payment amount, and complaint frequency are calculated. For example, the feature vector of power customer A is an average electricity consumption of 450 kWh, a total payment amount of 7200 yuan, and a complaint frequency of 2 times. The aggregated data set highlights the power customer behavior patterns and provides a basis for clustering. In one embodiment, the K-means clustering algorithm is used to classify the aggregated data set. Set K = 3 to divide the power customers into three groups: high electricity consumption and high complaints, low electricity consumption and no complaints, and medium electricity consumption and occasional complaints. For example, power customer A is assigned to the high electricity consumption and high complaints group, which reflects its behavior characteristics. It can be understood that the Apriori algorithm is used to analyze the power customer behavior patterns. The mining results show that high electricity consumption power customers are often associated with "voltage problem" complaints, and low electricity consumption power customers mostly have "no complaints".For example, continuous features such as electricity consumption, payment amount, and complaint frequency are classified into categories such as high / middle / low according to business rules, and the labels of power customer groups generated by K-means clustering are retained (such as high electricity consumption and high complaints). The group label of each power customer is combined with the discretized features to form a transaction (such as "high electricity consumption and high complaints + high electricity consumption + high complaint frequency"). The Apriori algorithm is used to set the minimum support, and high-frequency co-occurrence item sets that simultaneously contain power customer groups, electricity consumption characteristics, and complaint characteristics are extracted. Based on the frequent item sets, the minimum confidence is set, and strong association rules between power customer groups and electricity consumption / complaint patterns are screened (such as "high electricity consumption power customer group → high complaint frequency"). The rules are structured according to indicators such as support and confidence to form a unified power customer data set with a data structure containing the association relationships of power customer behaviors, and the result is obtained: 80% of the high electricity consumption power customer group has voltage-related complaints, revealing the association between electricity consumption and complaint types. The unified power customer data set with a data structure provides a basis for the enterprise to optimize services, such as strengthening voltage stability for high electricity consumption power customers. It should be noted that the above process forms a complete analysis chain through data cleaning, standardization, feature extraction, clustering, and association analysis. Each step supports each other to ensure data quality and analysis accuracy, and ultimately helps the enterprise understand power customer behaviors, optimize resource allocation, and improve service quality.

[0012] S102. For the unified power customer data set with a data structure, use the corr function of the pandas library to calculate the correlation coefficient between the features of the power customers, construct a Pearson correlation coefficient matrix, and generate the principal component feature vectors of the power customers by using the principal component analysis method based on the Pearson correlation coefficient matrix, so as to obtain the reduced-dimensional power customer feature set.

[0013] Obtain the power customer features from the unified power customer data set with a data structure, use the corr function of the pandas library to calculate the correlation coefficient between the features of the power customers, and construct a Pearson correlation coefficient matrix. The elements of the Pearson correlation coefficient matrix reflect the correlation coefficients between different features of the power customers. According to the preset correlation coefficient threshold, if the absolute value of the correlation coefficient corresponding to any two features exceeds the preset correlation coefficient threshold, they are determined as redundant features, and the feature pairs composed of the two features are marked. Use the var function of the numpy library to calculate the variance, retain the feature with the larger variance, eliminate the redundant features, and obtain the refined feature set. For the refined feature set, use the principal component analysis method to extract the principal component features and generate the principal component feature vectors to obtain the reduced-dimensional power customer feature set.

[0014] Exemplarily, power customer characteristics are obtained from a power customer dataset with unified data structure. The power customer dataset includes fields such as power customer identification, average power consumption, total payment amount, and complaint frequency. The corr function of the pandas library is used to calculate the correlation coefficients between the characteristics of power customers, construct a Pearson correlation coefficient matrix, and measure the linear correlation between characteristics. The range of the Pearson correlation coefficient is from -1 to 1, and an absolute value close to 1 indicates a strong correlation. For example, the dataset includes the average power consumption of power customer A as 450 kWh, the total payment amount as 7,200 yuan, and the complaint frequency as 2 times. Through calculation using the corr function, the correlation coefficient between the average power consumption and the total payment amount is 0.85, indicating a high correlation between the two. In a possible implementation, a preset correlation coefficient threshold is 0.8. If the absolute value of the correlation coefficient between any two characteristics exceeds 0.8, they are determined to be redundant characteristics. For example, the correlation coefficient of 0.85 between the average power consumption and the total payment amount exceeds the correlation coefficient threshold and is marked as a redundant characteristic pair. The var function of the numpy library is used to calculate the variances of the two. Suppose the variance of the average power consumption is 10,000, with the unit of kWh squared, and the variance of the total payment amount is 5,000, with the unit of yuan squared. The average power consumption with a larger variance is retained, and the total payment amount is removed to obtain a refined feature set. The refined dataset reduces redundant information and improves the efficiency of subsequent analysis. Specifically, for the refined feature set, the principal component analysis method is used to extract the principal components and generate the principal component feature vectors. The principal component analysis method combines the original features into new orthogonal features through linear transformation, retaining the main information of the data. For example, the refined feature set includes the average power consumption and the complaint frequency. After applying the principal component analysis method, two principal component feature vectors are generated. The first principal component feature vector may mainly reflect the power consumption behavior, and the second principal component feature vector reflects the complaint tendency. The dimension of the feature vector after dimensionality reduction is reduced, for example, from 10 dimensions to 2 dimensions, while retaining 90% of the information variance. Preferably, the power customer feature set after dimensionality reduction can be used for subsequent power customer behavior analysis. For example, the feature vector of power customer A after dimensionality reduction is [2.5, -0.3], reflecting its high power consumption and medium complaint tendency. Compared with the original high-dimensional data, the feature set after dimensionality reduction is more conducive to visualization and modeling, improving the analysis efficiency. It can be understood that the above process forms a logically rigorous feature optimization chain through correlation analysis, feature elimination, and dimensionality reduction processing. Correlation analysis identifies redundant characteristics, variance comparison ensures the retention of characteristics with large amounts of information, and the principal component analysis method further compresses the data dimension, jointly improving the data processing efficiency and analysis accuracy.

[0015] S103. Calculate the Euclidean norm of the principal component feature vector of the power customer based on the power customer feature set after dimensionality reduction to obtain the eigenvalue of the power customer, and determine whether the eigenvalue exceeds a preset threshold T1. If so, determine that the power customer is a high-value power customer.

[0016] Obtain the principal component feature vector of power customers from the dimension-reduced power customer feature set, and obtain the eigenvalue of power customers by calculating the Euclidean norm of the principal component feature vector of power customers. If the eigenvalue does not exceed the preset threshold T1, it is marked as an ordinary customer. If the eigenvalue exceeds the preset threshold T1, the customer is marked as a high-value power customer to obtain a high-value customer group.

[0017] Exemplarily, the dimension-reduced power customer feature set contains the principal component feature vector of power customers for further analysis of power customer value. Calculating the Euclidean norm is to take the square root of the sum of the squares of each component of the vector to obtain a single eigenvalue, quantifying the overall strength of power customer features. Specifically, the Euclidean norm of the feature vector [2.5, -0.3] of power customer A is calculated by , an eigenvalue of approximately 2.52 is obtained. This eigenvalue can be used to distinguish power customer types and simplify subsequent classification. In a possible implementation, a preset threshold T1 is used to divide ordinary power customers and high-value power customers. The setting of T1 is based on business requirements. For example, through historical data analysis, the median or average value of the eigenvalue is determined as a reference. Suppose T1 is set to 2.0. If the Euclidean norm of the power customer eigenvalue does not exceed 2.0, it is marked as an ordinary power customer; if it exceeds 2.0, it is marked as a high-value power customer. For example, the eigenvalue of power customer A, 2.52, exceeds T1 and is marked as a high-value power customer; the eigenvector of power customer B is [1.2, 0.4], and the calculated eigenvalue is approximately 1.26, which does not exceed T1 and is marked as an ordinary power customer. This classification method clearly distinguishes power customer groups by quantifying the feature strength, facilitating targeted management by enterprises. It should be noted that the calculation of the Euclidean norm is simple and efficient and is suitable for the comprehensive evaluation of high-dimensional feature vectors. Compared with other distance metrics, the Euclidean norm treats the weights of each component equally and is suitable for orthogonal features generated by the principal component analysis method. For example, the eigenvector of power customer C is [3.0, 1.0], and the eigenvalue is approximately 3.16, which significantly exceeds T1, indicating that its power consumption behavior or complaint tendency has outstanding performance and conforms to the characteristics of high-value power customers. Preferably, the classification based on eigenvalues can be refined in combination with business scenarios. For example, high-value power customers may be power customers with high power consumption or high complaint frequencies and need to be given priority attention. In one embodiment, the identification of high-value power customer groups can be further combined with power customer portraits. For example, the eigenvalue of power customer D is 3.5 and is marked as a high-value power customer. Its eigenvector [3.2, -0.8] shows high power consumption and low complaint tendency and is suitable for recommending customized power usage plans; the eigenvalue of power customer E is 2.8, and the eigenvector [1.5, 2.0] reflects medium power consumption and high complaint tendency, and its service experience needs to be optimized. It can be understood that this classification logic forms a tight chain from the eigenvector to the eigenvalue and then to the power customer grouping, ensuring that the classification results are interpretable and practical. In one embodiment, the threshold T1 can be dynamically adjusted to adapt to different business objectives. For example, at the initial stage, T1 is set to 2.0 to screen core high-value power customers, and later it is lowered to 1.8 to expand the target group. The eigenvalue of power customer F is 1.9 and is marked as an ordinary power customer at the initial stage, but may be converted to a high-value power customer due to the adjustment of T1 later. This flexibility improves the adaptability of the classification method. Exemplarily, the classified high-value power customer groups can be used for precision marketing or resource allocation to enhance the enterprise's response ability to key power customers.

[0018] S104. Obtain the power consumption and complaint frequency of the high-value power customers from the power customer dataset, and use time series analysis technology to calculate the change rates of the power consumption and complaint frequency within consecutive time windows to obtain the dynamic feature vectors of the high-value power customers.

[0019] Obtain the power consumption and complaint frequency of the high-value power customers from the power customer dataset, extract the time series data of the power consumption and complaint frequency, and generate a sequence dataset containing timestamps. For the sequence dataset, adopt the sliding window method, set the window size to 7 days and the sliding step to 1 day, calculate the change rate of the power consumption and complaint frequency within consecutive time windows, and obtain the change rate sequence. Through the change rate sequence, construct the dynamic feature vector of the high-value power customers.

[0020] Exemplarily, the electricity consumption and complaint frequency of high-value electricity customers are obtained from the electricity customer dataset to generate a time series dataset. The time series data is indexed by timestamps, recording the electricity consumption and the number of complaints of electricity customers at specific time points. For example, the electricity consumption data of high-value electricity customer A may be 1000 kWh on January 1, 2025, and 1050 kWh on January 2, and the complaint frequency is 0 times on January 1 and 1 time on January 2. The time series dataset is in daily units, containing the electricity consumption and complaint frequency of consecutive dates, forming structured sequence data. Such a dataset provides a basis for subsequent analysis. In a possible implementation, a sliding window method is used to process the time series data, with the window size set to 7 days and the step size to 1 day. The sliding window slides day by day, extracting the electricity consumption and complaint frequency data within 7 days and calculating their rates of change. The rate of change reflects the fluctuation trend of electricity consumption or complaint frequency within the time window. For example, the electricity consumption of electricity customer A increased from 1000 kWh to 1100 kWh from January 1 to January 7, with a rate of change of approximately 2.86% growth per day; the complaint frequency increased from 0 times to 2 times, with a rate of change of approximately 0.29 times per day. The rate of change sequence is in units of time windows, recording the rates of change of electricity consumption and complaint frequency for each window, forming a new feature sequence. It should be noted that the generation of the rate of change sequence needs to consider the stationarity of the data. The electricity consumption of electricity customers may be affected by seasons or holidays, and the complaint frequency may fluctuate due to service quality. The sliding window method smooths short-term fluctuations and highlights trend changes by fixing the window size. For example, the rate of change of electricity consumption of electricity customer B stabilized at 1.5% from January 8 to January 14, and the rate of change of complaint frequency was 0.1 times, reflecting that their electricity consumption behavior was stable but the complaint tendency increased slightly. This method ensures that the rate of change sequence can capture dynamic behavior patterns. Specifically, a dynamic feature vector of high-value electricity customers is constructed based on the rate of change sequence. The dynamic feature vector takes the rate of change of each time window as an element, describing the dynamic characteristics of electricity customers in terms of electricity consumption and complaint behavior. For example, the feature vector of electricity customer A in a certain window may be [2.86, 0.29], indicating an electricity consumption growth rate of 2.86% and a complaint frequency growth rate of 0.29 times; the feature vector of electricity customer B is [1.5, 0.1]. The dynamic feature vector reflects the time-dimensional characteristics of electricity customer behavior through multi-dimensional data integration, facilitating the subsequent analysis of the behavior trends of electricity customers. In one embodiment, the construction of the dynamic feature vector can adjust the window size in combination with business requirements. For example, the window size can be extended to 14 days to capture longer-term trends, or shortened to 3 days to highlight short-term fluctuations. The feature vector of electricity customer C in a 14-day window may be [1.8, 0.05], reflecting that their electricity consumption has a stable growth and few complaints. This flexibility enables the dynamic feature vector to adapt to different business scenarios. Preferably, the application of the dynamic feature vector can support predictive analysis of electricity customer behavior.For example, the high growth rate of electricity consumption of electricity customer A may indicate an increase in its future demand, and the enterprise can adjust the power supply plan in advance; the rising growth rate of the complaint frequency of electricity customer B may signal service problems and timely intervention is required. This method captures behavioral changes through dynamic features, improving the accuracy of electricity customer management. It can be understood that the above process forms a logically rigorous analysis chain from time series extraction to dynamic feature vector construction. The time series dataset ensures the time continuity of data, the sliding window method mines the change trend, and the dynamic feature vector integrates multi-dimensional information, providing actionable insights into the behavior of electricity customers for the enterprise. For example, the dynamic feature vector of electricity customer A can be used to predict its electricity consumption demand and optimize resource allocation; the low complaint feature of electricity customer C indicates its high satisfaction, and value-added services can be preferentially recommended. This method enhances the understanding and application ability of the behavior of high-value electricity customers through progressive analysis.

[0021] S105. For the dynamic feature vectors of high-value electricity customers, construct a multi-dimensional feature matrix of high-value electricity customers according to the time series, input the multi-dimensional feature matrix into a pre-trained long short-term memory network model, and output the behavior prediction results of high-value electricity customers, where the long short-term memory network model is constructed using a deep learning framework and trained using the historical multi-dimensional feature matrix as input.

[0022] Among them, the training of the long short-term memory network model includes: obtaining the historical dynamic feature vectors of historical high-value power customers, arranging the historical dynamic feature vectors in chronological order to form a historical multi-dimensional feature matrix including the time dimension, where each row of the historical multi-dimensional feature matrix corresponds to the feature vector of a time window, and each column corresponds to different time series features. The historical multi-dimensional feature matrix is divided into a training set, a validation set, and a test set in chronological order, where the training set is used for learning the parameters of the long short-term memory network model, the validation set is used for hyperparameter tuning and overfitting judgment, and the test set is used for final model performance evaluation. A long short-term memory network model is constructed using a deep learning framework. The model includes at least one hidden layer of the long short-term memory network model and one fully connected output layer of the long short-term memory network model. Among them, the hidden layer of the long short-term memory network model is used to capture the long-term dependencies in the time series, and the fully connected output layer of the long short-term memory network model is used to output the predicted customer behavior change trend. The training set is used to iteratively train the long short-term memory network model. The mean square error is used as the loss function, and the Adam adaptive optimization algorithm is selected as the optimizer. During the training process, the validation set is introduced to monitor the loss value of the model on non-training data in real time, and overfitting is avoided through the early stopping mechanism. When the loss of the validation set does not decrease within the preset number of rounds, the training is automatically terminated. The trained long short-term memory network model is used to predict the test set data to obtain the predicted values of the user behavior change trend, and the error between the predicted values and the actual values is calculated. The calculated error value is compared with the preset error threshold. If the error value is less than or equal to the preset error threshold, it is determined that the model prediction result is valid and regarded as the training completion. If the error value is greater than the preset error threshold, the model optimization process is triggered, the hyperparameters of the long short-term memory network model are adjusted, and the input of the training set, validation set, and test set is re-executed until the error value meets the preset conditions, where the hyperparameters include the number of neurons in the hidden layer, the time step, and the learning rate.

[0023] Exemplarily, historical dynamic feature vectors of high-value power customers are extracted from the power customer dataset. Assume that the feature vector of power customer A includes the electricity consumption change rate and the complaint frequency change rate. Each vector corresponds to a 7-day sliding window. For example, the vector from January 1 to January 7, 2024 is [2.5, 0.2], indicating a 2.5% daily average increase in electricity consumption and a 0.2 times daily average increase in complaint frequency. Arranged in chronological order, the feature vectors from January 1 to December 31, 2024 form a historical multi-dimensional feature matrix. Each row is a vector of a 7-day window, and each column is either the electricity consumption change rate or the complaint frequency change rate. The matrix size is 358 rows and 2 columns, and the time dimension is fully retained. In a possible implementation, the historical multi-dimensional feature matrix is divided into a training set, a validation set, and a test set in chronological order. Assume that the data from January 1 to October 31, 2024 is the training set, accounting for about 70%, that is, 250 rows, which is used for the long short-term memory network model to learn the power customer behavior pattern; from November 1 to November 30 is the validation set, accounting for about 10%, that is, 36 rows, which is used to tune the hyperparameters; from December 1 to December 31 is the test set, accounting for about 20%, that is, 72 rows, which is used to evaluate the model performance. This division ensures the continuity of the time series and avoids data leakage. Specifically, the construction of the long short-term memory network model is based on a deep learning framework. The model contains one hidden layer with 128 neurons, which captures the long-term dependencies of electricity consumption and complaint frequency, such as seasonal electricity consumption peaks or complaint concentration periods; the fully connected output layer outputs the predicted change rate of the next time window, such as [2.7, 0.3]. The hidden layer filters key information through the forget gate and the input gate, retaining the long-term trend, which is suitable for processing the complex time series patterns of power customer behavior. It should be noted that during the training process, the training set is used to iteratively optimize the model, and the mean squared error is used as the loss function to measure the deviation between the predicted value and the actual value. The Adam optimizer is suitable for the unstable fluctuations of time series data due to its adaptive learning rate adjustment. The validation set monitors the loss value in real time. For example, after 10 rounds of training, the validation set loss drops from 0.05 to 0.02. If it does not decrease for 5 consecutive rounds, the early stopping mechanism terminates the training to avoid overfitting. In one embodiment, the test set is used to predict the power customer behavior trend. Assume that the actual change rate of power customer A from December 1 to December 7, 2024 is [2.8, 0.25], and the model prediction is [2.75, 0.27], with an error of [0.05, 0.02]. The preset error threshold is 0.1, and the error meets the condition, so the model prediction is valid. If the error exceeds the standard, for example, the prediction is [3.0, 0.4], and the error is [0.2, 0.15], then the hyperparameters are adjusted, such as increasing the number of hidden layer neurons to 256 or reducing the learning rate to 0.0005, and retraining. Preferably, the hyperparameters are adjusted flexibly to meet the business requirements. For example, the time step is set to 7 to match the sliding window size to capture short-term trends; if the business needs to predict long-term trends, it can be increased to 14.The prediction results of Power Customer B show that the electricity consumption growth rate has increased from 1.5% to 2.0%, prompting the enterprise to plan power supply in advance. This method accurately captures the dynamic behavior of power customers through a multi-dimensional feature matrix and a long short-term memory network model. It can be understood that the above process forms a rigorous analysis chain from matrix construction to model prediction. The division of the training set, validation set, and test set ensures the generalization ability of the model. The hidden layer design mines the time series law, and the error evaluation and hyperparameter optimization guarantee the prediction accuracy. It can be understood that after the model training is completed, a multi-dimensional feature matrix of high-value power customers is constructed, and the multi-dimensional feature matrix is input into the long short-term memory network model to output the behavior prediction results of high-value power customers. For example, the low-complaint prediction result of Power Customer A indicates high satisfaction, and the enterprise can give priority to recommending value-added services to enhance the stickiness of power customers.

[0024] S106. Obtain the complaint records and interaction logs of high-value power customers from the power customer dataset, input the complaint records and interaction logs of high-value power customers into a preset sentiment analysis model for sentiment analysis to obtain the sentiment analysis results of high-value power customers, construct a sentiment analysis feature vector through the sentiment analysis results and the behavior prediction results, and input the sentiment analysis feature vector into a pre-constructed decision tree model for appeal complexity classification to obtain the appeal complexity classification result, where the appeal complexity classification result is one of the high-complexity appeal result, medium-complexity appeal result, and low-complexity appeal result. The sentiment analysis model is obtained by training the BERT-base model based on historical complaint records and interaction logs, and the decision tree model is trained using the sentiment analysis feature vector.

[0025] Obtain the complaint records and interaction logs corresponding to high-value customers in the power customer dataset with a unified data structure, merge the data according to the customer ID and timestamp to generate an original text dataset. Use regular expressions to remove special symbols and garbled characters in the original text, filter out irrelevant words through a stop word list, and output a standardized text dataset. Input the text data in the standardized text dataset into the Word2Vec model of the preset Gensim tool, fix the output dimension at 300 dimensions, and output 300-dimensional text feature vectors. Train a BERT-base sentiment analysis model based on historical complaint records and historical interaction logs, and input the 300-dimensional text feature vectors into the trained model to obtain the sentiment analysis results of high-value power customers. Standardize the behavior prediction results and sentiment analysis results through standard deviation standardization, and construct a sentiment analysis feature vector by combining the standardized sentiment analysis results and behavior prediction results. Input the sentiment analysis feature vector into a decision tree pre-trained according to historical sentiment analysis feature vectors to obtain the appeal complexity classification result including high-complexity appeal results, medium-complexity appeal results, and low-complexity appeal results.

[0026] Exemplarily, in the processing of power customer data, obtaining high-value customer complaint records and interaction logs requires ensuring the unity of the data structure. The complaint records include customer ID, timestamp, and complaint content. The interaction logs record the customer ID, timestamp, and interaction channels such as phone or online consultation. The data is merged and sorted by customer ID and timestamp to generate the original text dataset. The original dataset may contain thousands of records, such as 10 complaints and 50 interaction logs of power customer A. This merging preserves the temporal relationship and facilitates subsequent analysis of power customer behavior. In one possible implementation, regular expressions are used to clean the text data. Suppose the complaint text of power customer A contains special symbols "#￥%" and garbled characters "&%x". The regular expression can match and delete these characters, only retaining Chinese, English, and numbers. The stop word list filters out irrelevant words, such as "of" and "is". For example, the original text "The bill #￥% is incorrect &%x" becomes "The bill is incorrect" after cleaning. Each line of the standardized text dataset corresponds to a cleaned record. For example, 50 records of power customer A form a plain text list. This processing ensures that the text is suitable for subsequent feature extraction. Specifically, the text data in the standardized text dataset is input into the Word2Vec model of the preset Gensim tool, with the fixed output dimension of 300 dimensions, and 300-dimensional text feature vectors are output. Suppose the text dataset of power customer A contains 500 words. Word2Vec maps words such as "bill" and "complaint" to 300-dimensional vectors through the context. For example, a certain complaint of power customer A, "The bill is incorrect", generates a 300-dimensional vector. It can be understood that in the Word2Vec model, 300 dimensions is a commonly used dimension that balances computational efficiency and semantic expression ability (too low will lose information, too high will increase redundancy), which is sufficient to capture complex semantic relationships. This vector representation captures the semantic relationship and facilitates sentiment analysis. It should be noted that the BERT-base sentiment analysis model is trained based on historical complaints and interaction logs to identify the text sentiment tendency. The "The bill is abnormal" of power customer A is input into the model, and a negative sentiment score of 0.8 is output, reflecting dissatisfaction. The 300-dimensional vector is further input into BERT to enhance the accuracy of sentiment analysis. For example, behavioral prediction results such as a power consumption growth rate of 2.5% and sentiment analysis results such as a negative sentiment of 0.8 need to be standardized. Standard deviation standardization converts the data into a distribution with a mean of 0 and a standard deviation of 1. For example, the growth rate of 2.5% is standardized to 1.2, and the sentiment of 0.8 is standardized to 0.9, constructing a sentiment analysis feature vector [1.2, 0.9]. It can be understood that the sentiment analysis feature vector is input into the pre-trained decision tree, and the output is the classification of the complexity of the appeal. The decision tree learns the classification rules based on the historical vectors. For example, power customers with high power consumption growth and high negative sentiment may have high-complexity appeals. The vector [1.2, 0.9] of power customer A is predicted to have a medium-complexity appeal, prompting the enterprise to give priority to handling. In one possible implementation, the classification of the complexity of the appeal guides resource allocation.High-complexity requests require dedicated follow-up, medium-complexity requests are arranged for regular support, and low-complexity requests are resolved through self-service. For example, the low-complexity requests of power customer B are quickly resolved through online guidance, improving efficiency. This classification logic ensures the rationality of resource allocation and at the same time improves the satisfaction of power customers.

[0027] S107. Use a weighted scoring method to comprehensively score the electricity consumption, behavior dynamics, and classification results of the complexity of requests of high-value power customers, and determine the service priority classification results of high-value power customers based on the comprehensive score. Among them, if the comprehensive score is higher than the preset comprehensive score threshold T2, it is determined as a priority service object; otherwise, it is determined as an ordinary service object.

[0028] Obtain the electricity consumption data, behavior dynamics records, and classification results of the complexity of requests in the power customer dataset with a unified data structure, merge the data according to the customer ID and timestamp to generate a comprehensive dataset. In the comprehensive dataset, for the customer ID and timestamp, extract the values of electricity consumption, behavior dynamics, and complexity of requests. According to the pre-established weight distribution table, multiply the values of electricity consumption, behavior dynamics, and complexity of requests by the corresponding weights and add them up to obtain the comprehensive score. According to the preset comprehensive score threshold T2, if the comprehensive score is higher than T2, it is marked as a priority service object; if the comprehensive score is lower than or equal to T2, it is marked as an ordinary service object to obtain the service priority classification result.

[0029] Exemplarily, in the processing of high-value power customer data, to obtain power consumption data, behavior dynamics records, and appeal complexity classification results with unified data structures, it is necessary to ensure consistent data fields. The power consumption data includes customer IB, timestamp, monthly power consumption in kilowatt-hours; the behavior dynamics records include customer IB, timestamp, and behavior types such as payment frequency or consultation times; the appeal complexity classification results include customer IB, timestamp, and complexity levels such as high, medium, and low. The data is merged with customer IB and timestamp as keys to generate a comprehensive dataset. For example, power customer A consumed 500 kilowatt-hours on February 1, 2024, and consulted the electricity bill online on the same day, with the appeal complexity being medium. After merging, a unified record is formed. This merging method ensures the integrity of the time series and facilitates subsequent analysis. Specifically, to extract the power consumption, behavior dynamics, and appeal complexity values from the comprehensive dataset, clear mapping rules need to be defined. The power consumption directly takes the kilowatt-hour value, such as 500; the behavior dynamics can be quantified by the behavior frequency. For example, power customer A consults 2 times a month, which is mapped to a dynamics score of 2; the appeal complexity is assigned values according to the level, such as 3 for high, 2 for medium, and 1 for low. The medium complexity of power customer A is assigned a value of 2. This quantification method unifies different types of data into numerical values, facilitating weighted calculations. In a possible implementation, the weight distribution table is preset according to business requirements. The weight of power consumption is 0.4, reflecting the core service requirements; the weight of behavior dynamics is 0.3, reflecting the activity of power customers; the weight of appeal complexity is 0.3, highlighting the service priority. For example, for power customer A with a power consumption of 500, behavior dynamics of 2, and appeal complexity of 2, the comprehensive score is 500×0.4 + 2×0.3 + 2×0.3 = 201.2. This weighting method balances multi-dimensional data and comprehensively reflects the value of power customers. Preferably, the comprehensive score is compared with a preset threshold T2, and T2 is set to 150, reflecting the boundary of priority service. The score of power customer A, 201.2, is higher than T2 and is marked as a priority service object; if the score of power customer B is 120, which is lower than T2, it is marked as a general service object. Priority service objects can be assigned dedicated personnel to follow up, and general service objects are processed through standard processes. For example, power customer B is given priority response due to high power consumption and frequent consultations, and the electricity bill question is quickly resolved, improving satisfaction. It can be understood that the service priority classification results guide resource allocation. The high power consumption and complex appeals of priority service objects indicate their sensitivity to services and the need for quick response to maintain loyalty. For example, the consultation records of power customer A show concern about the transparency of electricity bills, and priority handling can reduce the risk of complaints. General service objects such as power customer B with low power consumption and simple appeals are suitable for self-service, optimizing resource utilization. In one embodiment, the comprehensive dataset supports dynamic adjustment of weights. For example, the weight of power consumption in the peak season can be increased to 0.5 to highlight the power consumption demand; the weight of behavior dynamics in the off-season can be increased to 0.4 to focus on power customer interaction. This flexibility adapts to business changes and ensures that the classification results meet actual needs.For example, the high electricity consumption of Power Customer A during the peak season further enhances its priority and obtains more efficient services. It should be noted that the quantification of behavioral dynamics records needs to be refined in combination with business scenarios. For example, a record of payment delay can reduce the dynamic score and reflect risks; frequent consultations increase the score and reflect activity. After comprehensively quantifying the on-time payment and multiple consultations of Power Customer A, the dynamic score is more accurate. This multi-faceted analysis ensures the comprehensiveness of the scoring system and the reliability of the classification results. For example, Power Customer A has an electricity consumption of 800 kWh, 3 monthly consultations, and a high complexity of demands, with a score of 800×0.4 + 3×0.3 + 3×0.3 = 321.8, far exceeding T2, and is marked as a priority service object. Its high electricity consumption and complex demands indicate that it requires special attention, and a quick response can enhance the trust of power customers. This multi-dimensional evaluation method generates an effective service priority classification result through weight balancing, which effectively guides service strategies, improves operational efficiency, and enhances the experience of power customers.

[0030] S108. If the service priority classification result of a high-value power customer is a priority service object, then a recommendation algorithm is used to generate a personalized service strategy for the high-value power customer. If the expected response time of the personalized service strategy is less than the preset threshold T3, it is determined as an executable strategy, and the final service optimization plan is obtained.

[0031] If the service priority classification result of a high-value power customer is a priority service object, then obtain the customer ID, electricity consumption, behavioral dynamics record, and demand complexity classification result data of the high-value power customer, and use SQL queries to extract customer needs and historical service records to generate a customer needs dataset. Based on the customer needs dataset, a user-based collaborative filtering algorithm is used to generate a personalized service strategy to obtain a preliminary strategy set. Extract the expected response time of each strategy from the preliminary strategy set, and use a linear regression model to quantify the expected response time to obtain a response time dataset. If the expected response time in the response time dataset is lower than the preset threshold T3, it is marked as an executable strategy, and SQL queries are used to generate an executable strategy set. According to the executable strategy set, a quicksort algorithm is used to sort the strategies by response efficiency to obtain an optimized service strategy sequence. Extract the highest-priority strategy from the optimized service strategy sequence, and use SQL queries to associate it with the customer ID and customer needs to generate the final service optimization plan.

[0032] Exemplarily, when obtaining the customer ID and attribute data of the priority service object from the service priority classification result, the required information can be extracted from the database through an SQL query. Assume that the customer relationship management system stores attributes such as customer ID, power consumption, behavior dynamics, and appeal complexity. The SQL query can be designed to filter the customer ID and related fields from the priority service object table. For example, the query statement can extract from the table priority_customers that the ID of power customer A is C001, the power consumption is 600 kWh, the number of queries is 4 times, and the appeal complexity is high. This query ensures accurate data extraction and provides a basis for subsequent analysis. In a possible implementation, when generating the power customer demand dataset, the SQL query can combine the power customer demand table and the service record table. Assume that the demand table records that power customer A submitted a bill dispute on April 5, 2025, and the service record table shows that the service durations of the past 3 services are 30 minutes, 45 minutes, and 20 minutes respectively. The query associates the two tables through the power customer ID and timestamp to form a dataset containing the power customer ID, demand type, and service duration. This association preserves the temporal relationship between demand and service, facilitating the analysis of power customer preferences. Specifically, based on the power customer demand dataset, a user-based collaborative filtering algorithm is used to generate personalized service strategies. Collaborative filtering recommends strategies suitable for C001 by analyzing the behaviors of similar power customers. For example, if power customer C002 has similar power consumption to C001 and consults frequently, the dedicated follow-up strategy of C002 can be recommended to C001. The initial strategy set includes dedicated service, self-service query guidance, etc. This algorithm utilizes the similarity between power customers to improve the pertinence of the strategies. Preferably, the estimated response time of each strategy is extracted from the initial strategy set. For example, dedicated service requires 1 hour and self-service query requires 0.5 hour. A linear regression model is used to quantify the response time to generate a response time dataset. Linear regression predicts the response time based on historical data. For example, the response time of dedicated service for C001 is 0.9 hour. The model considers factors such as service type and power customer activity to ensure accurate prediction. This quantification method provides a basis for strategy screening. It should be noted that if the response time is lower than the preset threshold T3, such as 1 hour, it is marked as an executable strategy. The SQL query filters the strategies with a response time lower than 1 hour from the database, such as the dedicated service and self-service query for C001, to generate an executable strategy set. This screening ensures the efficiency of the strategies and meets the needs of the priority service object. In an embodiment, the quicksort algorithm is used to sort the executable strategy set according to the response efficiency. Quicksort uses the response time as the key field and ranks the self-service query with 0.5 hour before the dedicated service with 0.9 hour to generate an optimized service strategy sequence. This sorting ensures that the most efficient strategy is executed first, improving the service response speed. It can be understood that the highest priority strategy is extracted from the optimized service strategy sequence, such as self-service query.SQL query associates it with customer ID and demand, such as binding C001's electricity bill detail demand with the self-service query strategy, and generates the final service optimization plan. The solution includes pushing a self-service query link to C001 to quickly resolve electricity bill questions. This association ensures that the strategy is accurately matched with power customer needs, improving service efficiency and power customer satisfaction.

[0033] Reference Figure 2 In an embodiment of the present application, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 2 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as customer complaint records and interaction logs. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the service optimization method for high-value power customers of any of the above embodiments is implemented.

[0034] Those skilled in the art will understand that Figure 2 The structure shown in is merely a block diagram of a portion of the structure related to the present application solution and does not constitute a limitation on the computer device to which the present application solution is applied.

[0035] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

Claims

1. A service optimization method for high-value power customers, characterized in that The method includes: Obtaining multi-dimensional data from the electricity consumption records, payment data, and complaint logs of power customers, and preprocessing the multi-dimensional data to obtain a power customer dataset with a unified data structure; for the power customer dataset with a unified data structure, using the corr function of the pandas library to calculate the correlation coefficients between the features of power customers, constructing a Pearson correlation coefficient matrix, and generating the principal component feature vectors of power customers using the principal component analysis method based on the Pearson correlation coefficient matrix to obtain a reduced-dimensional power customer feature set; according to the reduced-dimensional power customer feature set, calculating the Euclidean norm of the principal component feature vectors of power customers to obtain the eigenvalues of power customers, and determining whether the eigenvalues exceed a preset threshold T1. If so, determining that the power customer is a high-value power customer; obtaining the electricity consumption and complaint frequency of the high-value power customer from the power customer dataset, and using time series analysis technology to calculate the change rates of the electricity consumption and complaint frequency within a continuous time window to obtain the dynamic feature vectors of the high-value power customer; for the dynamic feature vectors of the high-value power customer, constructing a multi-dimensional feature matrix of the high-value power customer according to the time series, inputting the multi-dimensional feature matrix into a pre-trained long short-term memory network model, and outputting the behavior prediction results of the high-value power customer; obtaining the complaint records and interaction logs of the high-value power customer from the power customer dataset, inputting the complaint records and interaction logs of the high-value power customer into a preset sentiment analysis model for sentiment analysis to obtain the sentiment analysis results of the high-value power customer, constructing a sentiment analysis feature vector through the sentiment analysis results and the behavior prediction results, inputting the sentiment analysis feature vector into a pre-constructed decision tree model for appeal complexity classification to obtain the appeal complexity classification results; using a weighted scoring method to comprehensively score the electricity consumption, behavior dynamics, and appeal complexity classification results of high-value power customers, and determining the service priority classification results of high-value power customers based on the comprehensive score; if the service priority classification result of the high-value power customer is a priority service object, using a recommendation algorithm to generate a personalized service strategy for the high-value power customer. If the estimated response time of the personalized service strategy is less than a preset threshold T3, determining it as an executable strategy to obtain the final service optimization plan.

2. The method according to claim 1, characterized in that, The obtaining multi-dimensional data from the electricity consumption records, payment data, and complaint logs of power customers, and preprocessing the multi-dimensional data to obtain a power customer dataset with a unified data structure includes: Obtaining multi-dimensional data from the electricity consumption records, payment data, and complaint logs of power customers, and using the Pandas library to separate the structured fields and unstructured fields in the multi-dimensional data to obtain an initial dataset; Using the Pandas library to process the initial dataset. If missing values are detected, filling the numerical fields by the mean filling method and filling the categorical fields by the mode filling method to obtain the filled dataset; For the filled dataset, if it is detected that the value of a numerical field exceeds the preset numerical threshold range, then remove the outliers that exceed the numerical threshold range to obtain the cleaned dataset; Use StandardScaler to perform standardization processing on the cleaned dataset, and convert fields in different formats into a preset structure using a unified encoding rule to obtain the standardized dataset; Extract features including customer identification, electricity consumption, payment amount, and complaint frequency from the standardized dataset, and use the groupby function of the Pandas library to generate multi-dimensional feature vectors to obtain the aggregated dataset; Use the K-means clustering algorithm to classify the aggregated dataset, and divide customer groups according to the electricity consumption behavior and complaint behavior of power customers to obtain the classified power customer dataset; Use the Apriori algorithm to analyze the classified power customer dataset, determine the correlation relationship between the electricity consumption patterns and complaint patterns among power customer groups, and obtain the power customer dataset with a unified data structure.

3. The method according to claim 1, characterized in that, For the power customer dataset with a unified data structure, use the corr function of the pandas library to calculate the correlation coefficient between the features of power customers, construct a Pearson correlation coefficient matrix, and use the principal component analysis method based on the Pearson correlation coefficient matrix to generate the principal component feature vector of power customers to obtain the reduced-dimensional power customer feature set, including: Obtain the power customer features from the power customer dataset with a unified data structure, use the corr function of the pandas library to calculate the correlation coefficient between the features of power customers, and construct a Pearson correlation coefficient matrix. Among them, the elements of the Pearson correlation coefficient matrix reflect the correlation coefficients between different features of power customers; According to the preset correlation coefficient threshold, if the absolute value of the correlation coefficient corresponding to any two features exceeds the preset correlation coefficient threshold, it is determined as a redundant feature, mark the feature pair composed of the two features, use the var function of the numpy library to calculate the variance, retain the feature with the larger variance among them, and remove the redundant features to obtain the refined feature set; For the refined feature set, use the principal component analysis method to extract the principal component features and generate the principal component feature vector to obtain the reduced-dimensional power customer feature set.

4. The method according to claim 1, characterized in that, According to the reduced-dimensional power customer feature set, calculate the Euclidean norm of the principal component feature vector of the power customer to obtain the eigenvalue of the power customer, and determine whether the eigenvalue exceeds the preset threshold T1. If so, then determine that the power customer is a high-value power customer, including: Obtain the principal component feature vector of the power customer from the reduced-dimensional power customer feature set, and obtain the eigenvalue of the power customer by calculating the Euclidean norm of the principal component feature vector of the power customer; If the eigenvalue does not exceed the preset threshold T1, then mark it as an ordinary customer; If the eigenvalue exceeds the preset threshold T1, then mark the customer as a high-value power customer to obtain the high-value customer group.

5. The method according to claim 1, wherein Obtain the power consumption and complaint frequency of the high-value power customers from the power customer dataset, and use time series analysis technology to calculate the change rates of the power consumption and complaint frequency within consecutive time windows, so as to obtain the dynamic feature vectors of the high-value power customers, including: Obtain the power consumption and complaint frequency of the high-value power customers from the power customer dataset, extract the time series data of the power consumption and complaint frequency, and generate a sequence dataset containing timestamps; For the sequence dataset, adopt the sliding window method, set the window size to 7 days and the sliding step to 1 day, calculate the change rates of the power consumption and complaint frequency within consecutive time windows, and obtain the change rate sequence; Construct the dynamic feature vectors of the high-value power customers through the change rate sequence.

6. The method according to claim 1, wherein The training of the long short-term memory network model includes: Obtain the historical dynamic feature vectors of the historical high-value power customers, arrange the historical dynamic feature vectors in the order of time series to form a historical multi-dimensional feature matrix containing the time dimension, where each row of the historical multi-dimensional feature matrix corresponds to the feature vector of a time window, and each column corresponds to different time series features; Divide the historical multi-dimensional feature matrix into a training set, a validation set, and a test set in chronological order, where the training set is used for the parameter learning of the long short-term memory network model, the validation set is used for hyperparameter tuning and overfitting judgment, and the test set is used for the final model performance evaluation; Adopt a deep learning framework to construct a long short-term memory network model, which includes at least one hidden layer of the long short-term memory network model and one fully connected output layer of the long short-term memory network model. Among them, the hidden layer of the long short-term memory network model is used to capture the long-term dependencies in the time series, and the fully connected output layer of the long short-term memory network model is used to output the predicted customer behavior change trend; Use the training set to perform iterative training on the long short-term memory network model, adopt the mean square error as the loss function, select the Adam adaptive optimization algorithm as the optimizer, introduce the validation set in the training process to monitor the loss value of the model on the non-training data in real time, and avoid overfitting through the early stopping mechanism. Automatically terminate the training when the validation set loss does not decrease within the preset number of rounds; Use the trained long short-term memory network model to predict the test set data, obtain the predicted values of the user behavior change trend, and calculate the error between the predicted values and the actual values; Compare the calculated error value with the preset error threshold. If the error value is less than or equal to the preset error threshold, it is determined that the model prediction result is valid and regarded as the training is completed. If the error value is greater than the preset error threshold, trigger the model optimization process, adjust the hyperparameters of the long short-term memory network model, and re-execute the input of the training set, validation set, and test set until the error value meets the preset conditions, where the hyperparameters include the number of neurons in the hidden layer, the time step, and the learning rate.

7. The method according to claim 1, characterized in that, Obtain the complaint records and interaction logs of high-value power customers from the power customer dataset, input the complaint records and interaction logs of high-value power customers into a preset sentiment analysis model for sentiment analysis to obtain the sentiment analysis results of high-value power customers, construct a sentiment analysis feature vector through the sentiment analysis results and the behavior prediction results, input the sentiment analysis feature vector into a pre-constructed decision tree model for appeal complexity classification, and obtain the appeal complexity classification results, including: Obtain the complaint records and interaction logs corresponding to high-value customers in the power customer dataset with a unified data structure, merge the data according to the customer ID and timestamp to generate an original text dataset; Use regular expressions to remove special symbols and garbled characters in the original text, filter out irrelevant words through a stop word list, and output a standardized text dataset; Input the text data in the standardized text dataset into the Word2Vec model of the preset Gensim tool, fix the output dimension to 300 dimensions, and output 300-dimensional text feature vectors; Train a BERT-base sentiment analysis model based on historical complaint records and historical interaction logs, and input the 300-dimensional text feature vectors into the trained model to obtain the sentiment analysis results of high-value power customers; Perform standardization processing on the behavior prediction results and sentiment analysis results through standard deviation standardization, and construct a sentiment analysis feature vector by combining the standardized sentiment analysis results and behavior prediction results; According to the decision tree pre-trained based on historical sentiment analysis feature vectors, input the sentiment analysis feature vectors into the decision tree to obtain the appeal complexity classification results including high-complexity appeal results, medium-complexity appeal results, and low-complexity appeal results.

8. The method according to claim 1, characterized in that, The weighted scoring method is used to comprehensively score the electricity consumption, behavior dynamics, and appeal complexity classification results of high-value power customers, and the service priority classification results of high-value power customers are determined based on the comprehensive score, including: Obtain the electricity consumption data, behavior dynamics records, and appeal complexity classification results in the power customer dataset with a unified data structure, merge the data according to the customer ID and timestamp to generate a comprehensive dataset; In the comprehensive dataset, extract the values of electricity consumption, behavior dynamics, and appeal complexity for the customer ID and timestamp; According to the pre-established weight distribution table, multiply the values of electricity consumption, behavior dynamics, and appeal complexity by the corresponding weights and add them up to obtain the comprehensive score; According to the preset comprehensive score threshold T2, if the comprehensive score is higher than T2, it is marked as a priority service object, and if the comprehensive score is lower than or equal to T2, it is marked as an ordinary service object to obtain the service priority classification results.

9. The method according to claim 1, characterized in that If the service priority classification result of the high-value power customer is a priority service object, a recommendation algorithm is used to generate a personalized service strategy for the high-value power customer. If the expected response time of the personalized service strategy is less than the preset threshold T3, it is determined as an executable strategy to obtain the final service optimization plan, including: If the service priority classification result of high-value power customers is the priority service object, obtain the customer ID, power consumption, behavior dynamics record, and appeal complexity classification result data of high-value power customers, and use SQL query to extract customer needs and historical service records to generate a customer needs dataset; According to the customer needs dataset, use the user-based collaborative filtering algorithm to generate personalized service strategies to obtain a preliminary strategy set; Extract the estimated response time of each strategy from the preliminary strategy set, and use a linear regression model to quantify the estimated response time to obtain a response time dataset; If the estimated response time in the response time dataset is lower than the preset threshold T3, mark it as an executable strategy, and use SQL query to generate an executable strategy set; According to the executable strategy set, use the quicksort algorithm to sort the strategies according to the response efficiency to obtain an optimized service strategy sequence; Extract the highest-priority strategy from the optimized service strategy sequence, and use SQL query to associate it with the customer ID and customer needs to generate a final service optimization plan.

10. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 9.

Images