Big data analysis optimization method for power failure information service based on customer experience

Abstract

The invention relates to a customer experience-based big data analysis optimization method for a power failure information service, and the method comprises the steps: 1, collecting data, which relates to the main field of power failure information service complaints; 2, based on the data collected in the step 1, an improved KANO model is combined with a fuzzy clustering algorithm, a customer sensitivity thermodynamic matrix is constructed, customer features and demands are mined, and priorities of different group demands are accurately identified; step 3, establishing a user tagging model, including determining an objective function and a constraint condition of the user tagging model; and 4, establishing association mapping between the customer attribute features and the customer demand features by adopting a neural network model, and solving the customer tagging model. According to the invention, through the data-driven modeling and machine learning technology, the accuracy, initiative and inclusiveness of the power failure information service are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of power supply and distribution technology, and specifically relates to a big data analysis and optimization method for power outage information services based on customer experience. Background Technology

[0002] With the deepening of the informatization and intelligentization of smart distribution networks, the demand for large-scale data interaction in the power system is experiencing explosive growth. This trend is mainly driven by three factors: First, the large-scale deployment of intelligent devices such as smart meters, fault indicators, and distributed sensing terminals generates massive amounts of operational data streams every second; second, the deep integration of the three core businesses of marketing, distribution, and dispatch is accelerating the development of the integrated operation, distribution, and dispatch process, urgently requiring the breaking down of traditional business barriers to achieve cross-system collaborative operation; third, the continuously rising expectations of power customers for power supply reliability and service quality are forcing power grid companies to restructure their service chains. Against this backdrop, power outage information dissemination, as an advanced application scenario for the integration of operation, distribution, and dispatch data, has evolved into a key carrier for enhancing the value of data services. Its strategic significance lies not only in the rapid response to fault events, but also in the optimization of the entire process, including intelligent prediction of outage scope, dynamic scheduling of repair resources, and closed-loop management of customer notifications, driving the strategic transformation of distribution network operation from "passive repair" to "proactive prevention." This transformation will directly promote the improvement of the level of refined distribution network management and the optimization of customer service quality, laying a solid foundation for building a modern smart distribution network system.

[0003] In recent years, with the accelerated advancement of the intelligent and digital transformation of power systems, research on power outage information management is undergoing profound changes. Traditional power outage information management mainly focuses on post-event statistics and analysis, using manual recording and report summarization for fault statistics. This method has inherent defects such as data lag and insufficient accuracy. Modern power systems place higher demands on power outage information management, requiring a shift from passive response to proactive prevention, from post-event statistics to real-time monitoring, and from extensive management to precise service. Against this backdrop, refined and dynamic management methods based on big data analysis have emerged. These methods fully utilize the massive amounts of data collected by smart meters, fault indicators, and SCADA systems, combined with data mining and machine learning techniques, to achieve rapid identification, accurate location, and impact assessment of power outage events. Of particular note is the significant reduction in the collection and transmission latency of power outage information with the application of new technologies such as 5G communication and the Internet of Things, enabling power outage management applications with high real-time requirements. This transformation not only improves the efficiency of power grid fault handling but also lays a solid foundation for the optimization and innovation of power outage information services.

[0004] Digital technology provides comprehensive underlying support for power outage information management and has become an indispensable key technology in the digital transformation of power grid companies. Traditional power outage information management mainly relies on manual reporting, a model with many drawbacks: long information transmission chains, slow response times, and difficulty in guaranteeing data accuracy. Especially during large-scale power outages, manual reporting systems are often overwhelmed. In contrast, real-time reporting methods based on smart meters and communication technologies demonstrate significant advantages: through the proactive reporting function of smart meters, power outage detection can be completed within seconds; with the help of high-speed communication networks, power outage information can be transmitted to the main station system in real time; and by utilizing the spatial analysis capabilities of GIS platforms, accurate mapping of the outage area can be achieved. These technological advancements have improved the efficiency of power outage information collection by several orders of magnitude, saving valuable time for subsequent information services and fault handling.

[0005] The demand for power outage information services from electricity customers is undergoing profound changes, shifting from a traditional passive response to a proactive awareness. This shift is mainly reflected in three aspects: First, customers expect significantly higher timeliness in obtaining power outage information, hoping to be informed as soon as possible; second, customers have stricter requirements for the accuracy of information, including the cause of the outage, the scope of impact, and the estimated restoration time; third, customers expect personalized information services, such as the ability to choose notification methods and customize information content. However, traditional service models still have significant shortcomings: planned power outage notifications mainly rely on manual delivery, which is not only labor-intensive and costly but also suffers from low coverage and a lack of feedback channels; during fault-based power outages, due to information asymmetry, customers are often in an information blind spot, easily leading to anxiety and dissatisfaction. These problems are particularly prominent during extreme weather events or large-scale power outages. Therefore, building intelligent customer interaction systems has become a research hotspot in the industry. These systems, through automated and intelligent technologies, aim to achieve accurate delivery and two-way interaction of power outage information, fundamentally improving the customer service experience.

[0006] Despite significant progress in intelligent customer interaction systems, there are still obvious shortcomings in meeting diverse needs. Services for special groups are another area urgently needing improvement. These include the elderly, visually impaired, and hearing impaired individuals, who have specific information service needs, but existing systems often lack targeted service solutions. Furthermore, different industries have significantly different focuses regarding power outage information: residential users are more concerned about the duration of the outage, commercial users are concerned about economic losses, and industrial users prioritize power supply reliability. These differentiated needs place higher demands on power outage information service systems, requiring greater adaptability and flexibility. Solving these problems requires not only technological innovation but also a fundamental shift in service philosophy, truly establishing a customer-centric service mindset.

[0007] At the technical implementation level, existing power outage information service systems still face key challenges such as insufficient algorithm generalization capabilities. Due to the vast coverage area and diverse customer types of my country's power grid, a single algorithm model is insufficient to adapt to all scenarios. For example, the electricity consumption characteristics of urban and rural areas differ significantly, and the performance of the same algorithm may vary considerably in different regions. Furthermore, with the advancement of new power system construction and the continuous integration of new elements such as distributed power sources and electric vehicles, the operating characteristics of the power grid are constantly changing, placing higher demands on the adaptability and robustness of algorithms. Overcoming these technical bottlenecks requires systematic innovation across multiple dimensions, including data governance, model design, and algorithm training. Summary of the Invention

[0008] This invention addresses the shortcomings of existing technologies by proposing a big data analysis and optimization method for power outage information services based on customer experience.

[0009] The above-mentioned objective of this invention is achieved through the following technical solution: A big data analysis and optimization method for power outage information services based on customer experience includes the following steps: Step 1: Collect data, focusing on the main areas where complaints about power outage information services arise; Step 2: Based on the data collected in Step 1, an improved KANO model combined with a fuzzy clustering algorithm is used to construct a customer sensitivity heat map, and to mine customer characteristics and needs, so as to accurately identify the priority of the needs of different groups. Step 3: Establish a user tagging model, including determining the objective function and constraints of the user tagging model; Step 4: Use a neural network model to establish the correlation mapping between customer attribute features and customer demand features, and solve the customer tagging model.

[0010] Furthermore, the data collected in step 1 includes: 95598, 12398, 12345, hotline work order information, survey questionnaire information, improved KANO model parameters, and user tagging model parameters.

[0011] Furthermore, step 2 includes: Step 2.1: First, design a bipolar Likert 5-point scale with positive and negative questions for each service item; stratify sampling by customer type, with no fewer than 200 valid questionnaires per type, ensuring a confidence level of ≥95%; Step 2.2: Then, when using the improved KANO model to analyze the structured questionnaire survey, the questionnaire scores are converted into triangular fuzzy numbers, and the membership degree between each service and the KANO category is calculated by fuzzy clustering algorithm. Step 2.3: Finally, output the sensitivity heat map results for each customer type.

[0012] Furthermore, in step 2.2, converting questionnaire scores into triangular fuzzy numbers is achieved by quantifying the fuzziness of customer needs through membership functions, including: Fuzzification processing is performed: triangular fuzzy numbers are used to represent questionnaire scores, and membership functions are defined. : (1) In the formula: The questionnaire is scored on a scale of 1 to 5. , , Let these be the lower bound, most likely value, and upper bound of the fuzzy number; Based on the fuzzy KANO determination rule, the preliminary membership degree of the questionnaire scores to each need category is calculated; let... In the fuzzy reasoning rule, all conclusions are the first... Demand Category The rule set is Then the first Customer type for the first Service rating belong Preliminary membership Determined by the following formula: (2) In the formula: and The positive and negative fuzzy scores are obtained from formula (1), respectively. and The first The positive and negative fuzzy language values ​​corresponding to each rule; Sensitivity quantification: Determine the sensitivity intensity by calculating the fuzzy proximity of the demand classification. : (3) In the formula, For the first Customer type for the first This service belongs to the first The degree of membership of a demand. For the first Customer type for the first Questionnaire rating for the service For the first The weight of each demand category, K This represents the total number of demand categories.

[0013] Furthermore, in step 2.2, the formula for calculating the membership degree between each service and the KANO category using the fuzzy clustering algorithm is as follows: (4) In the formula, This is the sum of weighted distances within the class. For the sample Cluster centers membership degree v k For the first k Cluster center vectors, For fuzzy index, For the sample size, The total number of clusters and numerically equal to K same.

[0014] Moreover, in step 3, the objective function consists of prediction error, regularization term, and business logic penalty term, as shown in equation (5): (5) In the formula: For the total loss function, This is the weight matrix of the neural network. For the neural network bias vector, For customers The probability vector of actual demand; The demand probability vector predicted by the neural network. Customer type For services Sensitivity; , These are the regularization coefficient and the business logic penalty coefficient, respectively. M Total number of service items; The Frobenius norm of the weight matrix. The L2 norm of the bias vector. This is the sensitivity scaling factor. For customers predicted by the model For services The probability of demand.

[0015] Furthermore, in step 3, the constraints include: Feature standardization constraints: The attributes and requirements need to be standardized to the [0,1] interval, as shown in equations (6) and (7): (6) (7) In the formula: For customers Attribute feature vectors; The total number of features; For customers The demand feature vector; Total number of service demand types; Probability normalization constraints: The sum of the probabilities of each customer's demand for each service item is 1, and the probabilities are non-negative, as shown in equations (8) and (9): (8) (9) Sensitivity threshold constraint: The predicted probability is not less than a specified proportion of the service sensitivity corresponding to the customer type, as shown in equation (10): (10) In the formula: For type The customer base; Business logic constraints: High-consumption customers are predicted to have higher electricity cost sensitivity, while medium-consumption customers are concerned about electricity costs but have a certain tolerance. A balanced forecast is needed. Electricity costs account for a small proportion of their total electricity consumption, and their electricity cost sensitivity is relatively low, as shown in equation (11). (11) In the formula: To represent the electricity consumption levels, low electricity consumption is assigned a value of 0, medium electricity consumption is assigned a value of 0.5, and high electricity consumption is assigned a value of 1. Predicted demand probability for electricity cost early warning services.

[0016] Furthermore, in step 4, during training, the neural network uses the gradient backpropagation algorithm to adjust the weight network of the hidden layers, forming a mapping relationship between the input attribute features and the output demand features. The mathematical definitions of the input feature matrix and the output demand matrix are shown in equations (12) and (13): (12) (13) In the formula: For the input feature matrix, For the number of customer samples, The total number of attribute features for each customer. No. The first customer Each attribute feature value, To output the demand matrix, The total number of service demand types, For the first The first customer The probability of actual demand for each service item.

[0017] The advantages and positive effects of this invention are as follows: 1. This invention proposes an intelligent customer demand prediction and optimization method based on big data analysis. Through data-driven modeling and machine learning techniques, it significantly improves the accuracy, proactivity, and inclusiveness of power outage information services. In demand prediction, an improved Kano model combined with a fuzzy clustering algorithm is used to construct a customer sensitivity heatmap matrix, accurately identifying the demand priorities of different groups. A neural network prediction model then establishes a mapping relationship between customer attributes and demand characteristics, enabling accurate prediction and personalized notification pushes for highly sensitive customers, improving the coverage of planned power outage notifications in pilot areas while reducing complaint rates. In service process optimization, this invention establishes a closed-loop feedback mechanism, integrating multi-channel feedback data and dynamically adjusting notification strategies, automatically pushing repair progress and estimated recovery time during power outages. Regarding differentiated services, for key users such as hospitals and schools, the system provides priority services after identification through the heatmap matrix, significantly improving service coverage for special groups. In terms of power grid digital transformation, customer demand heatmap results can guide power grid transformation decisions, such as adding backup lines in areas with high complaint rates, reducing power outages at the source. Simultaneously, automated demand prediction significantly reduces labor costs, providing strong support for the digital transformation of power grid companies.

[0018] 2. Through an innovative architecture of big data analysis, machine learning, and closed-loop feedback, this invention has successfully achieved a transformation from passive response to proactive service, from a single model to personalized customization, and from experience-driven to data-driven approaches. This not only significantly improves customer experience but also provides a replicable solution for the digital transformation of the public utilities sector. Attached Figure Description

[0019] Figure 1 This is a flowchart of the big data analysis and optimization method for power outage information services based on customer experience, as proposed in this invention. Detailed Implementation

[0020] The structure of the present invention will be further described below with reference to the accompanying drawings and embodiments. It should be noted that these embodiments are descriptive and not limiting.

[0021] For a big data analysis and optimization method for power outage information services based on customer experience, please refer to [link / reference]. Figure 1 Its inventive point is that it includes the following steps: Step 1: Data Collection. The collected data covers the main areas where complaints about power outage information services arise. The collected data primarily includes: various work order information, survey questionnaire information, and model parameters, among which: The various types of work order information include: 95598 complaint work order content, 12345 complaint work order content, 12398 complaint work order content, and hotline complaint work order content.

[0022] The survey questionnaire information includes: the number of valid questionnaires, user electricity usage type, user industry distribution, sensitive content, and scores for each questionnaire.

[0023] Model parameters include improved KANO model parameters and user-labeled model parameters: The improved KANO model parameters include: the lower limit, most likely value, and upper limit of the fuzzy number; The parameters of the user tagging model include: sensitivity scaling factor, regularization factor, and business logic penalty factor.

[0024] Step 2: Based on the data input in Step 1, an improved Kano model combined with a fuzzy clustering algorithm is used to construct a customer sensitivity heatmap matrix. This matrix is ​​then used to mine customer characteristics and needs, accurately identifying the priority of needs for different groups (the customer sensitivity heatmap matrix serves as a demand probability matrix input into the subsequent user labeling model; the matrix reflects the user's sensitivity to different services, i.e., the priority of their needs). Specifically: Starting with data sources such as 95598 service orders and business hall service orders, this invention aims to identify the influencing factors of customer complaints from a wealth of complex data, enabling proactive preventative measures and service preparation to improve work quality, efficiency, and service levels. Based on a big data analytics platform, complaint order analysis can clarify specific improvement targets, facilitate specialized analysis, and formulate targeted improvement measures. This invention will focus on the main areas where complaints arise from power outage information services.

[0025] Online questionnaires were distributed to customers via electricity bills, WeChat official account pushes, Weibo links, and the mobile power app. After invalid questionnaires were removed, the remaining valid questionnaires were analyzed using Hadoop big data tools to determine the industry distribution and electricity consumption patterns of the sampled data.

[0026] This paper utilizes an improved Kano model combined with a structured questionnaire survey. Based on data such as customer type and scores for various sensitive items within the questionnaire, an analysis is conducted to establish a model of potential sensitive customers. This model includes customer sensitivity models categorized by different customer types and by different sensitive items. Different customer types include large industrial, general commercial and industrial, residential, and agricultural sectors; different sensitive items include electricity bills, power outages, and emergency repairs, and can be adjusted as needed. The traditional Kano model categorizes needs into five types—basic, expected, exciting, indifferent, and negative—using a binary questionnaire (positive / negative questions), but suffers from limitations due to fuzzy classification boundaries. This invention introduces fuzzy set theory, quantifying the fuzziness of customer needs through membership functions. Specific improvements are as follows: 1. Fuzzification: Triangular fuzzy numbers are used to represent questionnaire scores, and membership functions are defined. : (1) In the formula: The questionnaire is scored on a scale of 1 to 5. , , Let be the lower bound, most likely value, and upper bound of the fuzzy number.

[0027] Based on the fuzzy KANO determination rule, the preliminary membership degree of the questionnaire scores to each need category is calculated. Let... In the fuzzy reasoning rule, all conclusions are the first... Demand Category The rule set is Then the first Customer type for the first Service rating belong Preliminary membership Determined by the following formula: (2) In the formula: and The positive and negative fuzzy scores are obtained from formula (1), respectively. and The first The positive and negative fuzzy language values ​​corresponding to each rule.

[0028] 2. Sensitivity Measurement: Determine the sensitivity intensity by calculating the fuzzy proximity of the demand classification. : (3) In the formula, For the first Customer type for the first This service belongs to the first The degree of membership of a demand. For the first Customer type for the first Questionnaire rating for the service For the first The weight of each demand category, K This represents the total number of demand categories.

[0029] First, a bipolar Likert 5-point scale (1-5 points) was designed, with each service having both positive (when provided) and negative (when absent) questions. Stratified sampling was used by customer type, with at least 200 valid questionnaires per type, ensuring a confidence level ≥95%. Then, when using an improved KANO model combined with structured questionnaires, the questionnaire scores were converted into triangular fuzzy numbers, and the membership degree between each service and the KANO category was calculated using a fuzzy clustering algorithm. (4) In the formula, This is the sum of weighted distances within the class. For the sample Cluster centers membership degree v k For the first k Cluster center vectors, The fuzzy index is usually set to 2. For the sample size, The total number of clusters and numerically equal to k same.

[0030] Finally, the sensitivity heatmap matrix for each customer type is output, which is then used as the demand probability matrix to input into the subsequent user tagging model.

[0031] Step 3: Establish a user tagging model, including determining the objective function and constraints of the user tagging model. 1. Objective function Customer attribute features and customer demand samples are numerically processed to construct customer attribute feature vectors and customer demand feature vectors. Each feature value in the attribute feature vector corresponds to the numerical value or category of each attribute feature, and each feature value in the demand feature vector corresponds to a type of service demand and its demand level. The goal of the user tagging model is to achieve accurate prediction of customer demand probability through the mapping relationship between customer attributes and demands. Its objective function consists of prediction error, regularization term, and business logic penalty term, as shown in equation (5): (5) In the formula: For the total loss function, This is the weight matrix of the neural network. For the neural network bias vector, For customers The probability vector of actual demand; The demand probability vector predicted by the neural network. Customer type For services Sensitivity; , These are the regularization coefficient and the business logic penalty coefficient, respectively. M Total number of service items; The Frobenius norm of the weight matrix. The L2 norm of the bias vector. This is the sensitivity scaling factor. For customers predicted by the model For services The probability of demand.

[0032] 2. Constraints (1) Feature standardization constraints The attributes and requirements need to be standardized to the [0,1] interval, as shown in equations (6) and (7).

[0033] (6) (7) In the formula: For customers Attribute feature vectors; The total number of features; For customers The demand feature vector; This represents the total number of service demand types.

[0034] (2) Probability normalization constraint Ensure that the sum of the probabilities of each customer’s demand for each service item is 1 and the probabilities are non-negative, as shown in equations (8) and (9).

[0035] (8) (9) (3) Sensitivity threshold constraint The predicted probability must be no less than a specified proportion of the service sensitivity corresponding to the customer type, as shown in Equation (10).

[0036] (10) In the formula: For type The customer base.

[0037] (4) Business logic constraints High-consumption customers are predicted to have higher electricity cost sensitivity, while medium-consumption customers are concerned about electricity costs but have a certain tolerance. A balanced forecast is needed. Electricity costs account for a small proportion of their total electricity consumption, and their electricity cost sensitivity is relatively low, as shown in equation (11). (11) In the formula: To represent the electricity consumption levels, low electricity consumption is assigned a value of 0, medium electricity consumption is assigned a value of 0.5, and high electricity consumption is assigned a value of 1. Predicted demand probability for electricity cost early warning services.

[0038] Step 4: Use a neural network model to establish the correlation mapping between customer attribute features and customer demand features, thereby solving the customer tagging model. The attribute feature vectors of each sample are combined to form an attribute feature matrix, which is used as input to the neural network model. The customer demand feature vectors with known service requirements are combined to form a sensitivity heatmap matrix, used for model training. The neural network model establishes a mapping between customer attribute features and customer demand features. The neural network model is trained using customer samples with known service requirements. After training, the neural network automatically extracts and combines the input customer attribute features and calculates the probability of customer demand for each service item.

[0039] During training, the neural network uses the gradient backpropagation algorithm (BP algorithm) to adjust the weight network of the hidden layer, thereby forming a mapping relationship between the input attribute features and the output demand features. The mathematical definitions of the input feature matrix and the output demand matrix are shown in equations (12) and (13).

[0040] (12) (13) In the formula: For the input feature matrix, For the number of customer samples, The total number of attribute features for each customer. No. The first customer Each attribute feature value, To output the demand matrix, The total number of service demand types, For the first The first customer The probability of actual demand for each service item.

[0041] Neural networks update weights through backpropagation (BP algorithm), and the specific steps are as follows: 1. Forward propagation: Input layer, hidden layer, output layer, as shown in equations (14)-(16): (14) (15) (16) In the formula: For the first The weighted input vector of the layer; For the first Layer weight matrix; For the first The activation value output vector of the layer; No. Layer bias vector; For activation functions; This represents the total number of layers in the neural network.

[0042] 2. Loss Calculation: (17) In the formula: This represents the total loss function value. The total number of training samples; For the first The true demand probability vector of each sample; For the first The input vector of each sample in the output layer; This is the activation function for the output layer.

[0043] 3. Backpropagation gradient: (18) (19) In the formula: For the loss function on the th The gradient of the layer weights; For the first The sample at the th The error term vector of the layer; For Hadamard product (element-by-element multiplication); This is the derivative of the activation function.

[0044] 4. Parameter update: (20) (twenty one) In the formula: The learning rate; For the loss function on the th The gradient of layer bias.

[0045] 5. Validate the user tagging model. The user tagging model in this invention uses scikit-learn's built-in model performance validation tool to calculate prediction accuracy. Specifically: The mesh search parameter space consists of the hidden layer structure parameter mesh (hidden_layer_sizes) and the regularization coefficient mesh. The training process is terminated if the validation set loss does not decrease for 10 consecutive training iterations; the hit rate calculation function is: (twenty two) In the formula: This is the model hit rate, used to predict accuracy. This represents the number of samples in the test set. For indicator functions; Serving index for the highest probability predicted by the model The index serves the highest probability of actual demand.

[0046] Although embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A big data analytics optimization method for customer experience based outage information service, characterized in that: Comprising the following steps: Step 1, collect data, the collected data involves the main field of power information service complaint; Step 2, based on the data collected in step 1, using improved KANO model combined with fuzzy clustering algorithm, build customer sensitivity heat matrix, carry out customer characteristics and demand mining, accurately identify the priority of different group demand; Step 3, establish user labeling model, including determining the objective function and constraint condition of user labeling model; Step 4, using neural network model to establish the correlation mapping between customer attribute characteristics and customer demand characteristics, realize the solution of customer labeling model.

2. The big data analytics optimization method for customer experience based power outage information service of claim 1, wherein: In step 1, the collected data includes: 95598, 12398, 12345, hotline work order information, questionnaire information, improved KANO model parameters and user labeling model parameters.

3. The big data analytics optimization method for customer experience based power outage information service of claim 1, wherein, Step 2 includes: Step 2.1, first design bipolar Likert 5 level scale question, each service in the question corresponds to positive and negative question; stratified sampling according to customer type, not less than 200 valid questionnaires for each type, ensure that the confidence is greater than or equal to 95%; Step 2.2, then in the analysis of improved KANO model combined with structured questionnaire survey, convert the questionnaire score into triangular fuzzy number, and calculate the membership degree of each service and KANO category through fuzzy clustering algorithm; Step 2.3, finally output the sensitivity heat result of each customer type.

4. The big data analytics optimization method for customer experience based power outage information service of claim 1, wherein, In step 2.2, the conversion of questionnaire score into triangular fuzzy number is realized by quantifying the fuzziness of customer demand through membership function, including: Fuzzification: Triangular fuzzy numbers are used to represent the questionnaire scores, and membership functions are defined : (1) wherein: wherein is the questionnaire score, the questionnaire score is 1-5, , , is the lower bound, the most probable value, and the upper bound of the fuzzy number; Based on the fuzzy KANO determination rule, the preliminary membership degree of the questionnaire scores to each need category is calculated; let... In the fuzzy reasoning rule, all conclusions are the first... Demand Category The rule set is Then the first Customer type for the first Service rating belong Preliminary membership Determined by the following formula: (2) In the formula: and are the positive and negative problem fuzzy scores obtained from formula (1), respectively; and are the positive and negative fuzzy linguistic values corresponding to the first rule, respectively Sensitivity quantification: Determine the strength of sensitivity by calculating the fuzzy closeness degree of demand classification : (3) wherein, is the number of the first category of customers to the first service belongs to the first demand category, is the number of the first category of customers to the first service, is the weight of the first demand category, K is the total number of demand categories.

5. The big data analytics optimization method for customer experience based power outage information service as claimed in claim 1 wherein, In step 2.2, the membership degree formula of each service and KANO category calculated by fuzzy clustering algorithm is: (4) wherein is the sum of the within-class weighted distances, is the number of samples is the membership of the cluster center to the cluster, v k is the number of cluster centers, k is the cluster center vector, is the fuzziness index, is the number of samples, is the total number of clusters and is numerically the same as K .

6. The big data analytics optimization method for customer experience based power outage information service as claimed in claim 1 wherein, In step 3, the objective function is composed of prediction error, regularization term and business logic penalty term, as shown in formula (5): (5) wherein: is the total loss function, is the neural network weight matrix, is the neural network bias vector, is the true demand probability vector of the customer ; is the neural network predicted demand probability vector, is the sensitivity of the customer type to the service ; , are the regularization coefficient and the business logic penalty coefficient, respectively, M is the total number of service items; is the Frobenius norm of the weight matrix, is the L2 norm of the bias vector, is the sensitivity scaling coefficient, is the model predicted demand probability of the customer to the service .

7. The big data analytics optimization method for customer experience based power outage information service as claimed in claim 1 wherein, In step 3, the constraint conditions include: Characteristic standardization constraint: The attribute and demand characteristics need to be standardized to the interval [0, 1], as shown in formula (6) and (7): (6) (7) In the formula: is the attribute feature vector of the customer; is the total number of features; is the demand feature vector of the customer; is the total number of service demand types;​​ Probability normalization constraint: The sum of the demand probability of each customer for each service item is 1, and the probability is non-negative, as shown in formula (8) and (9): (8) (9) Sensitivity threshold constraint: The predicted probability is not less than the specified proportion of customer type corresponding service sensitivity, as shown in formula (10): (10) In the formulae: is a set of customers of the type ; Business logic constraint: High power consumption customers are predicted to have high electricity cost sensitivity, medium power consumption customers pay attention to electricity cost but have certain tolerance, and the predicted electricity cost proportion is small, so the electricity cost sensitivity is relatively low, as shown in formula (11): (11) In the formula: is the electricity consumption level characteristic, 0 for low electricity consumption, 0.5 for medium electricity consumption, and 1 for high electricity consumption, is the predicted demand probability of electricity fee warning service.

8. The big data analytics optimization method for customer experience based power outage information service of claim 1, wherein, In step 4, the neural network uses gradient back propagation algorithm to adjust the weight network of hidden layer during training, forming the mapping relationship between input attribute characteristics and output demand characteristics, and the mathematical definition of input feature matrix and output demand matrix, as shown in formula (12) and (13): (12) (13) where: is the input feature matrix, is the number of customer samples, is the total number of attribute features for each customer, is the attribute feature value of the customer, is the output demand matrix, is the total number of service demand types, is the true demand probability of the item service of the customer.

Images