Big data mining method, system and device for analysis and prediction of electric power service hotspot events and medium

Abstract

The invention discloses a big data mining method, system, equipment and medium for electric power service hotspot event analysis and prediction, and belongs to the technical field of electric power customer services, and the method comprises the steps: obtaining customer service data of a plurality of electric power service interaction systems, and constructing an electric power service risk comprehensive data set containing customer appeals and service behaviors; performing joint analysis of voiceprint features and text semantics on the client interaction content, and generating a client emotion score of each service event; constructing a dynamic heterogeneous map, analyzing a propagation path of the risk event in a space-time dimension by using a map neural network model, and generating a risk diffusion prediction result of each service event; calculating a comprehensive risk index of each service event; and when the comprehensive risk index exceeds a preset multi-level threshold, generating an early warning work order of a corresponding level, and triggering a message notification to a dispatcher of the corresponding level according to the risk area and the event type indicated by the comprehensive risk index. According to the invention, automation of risk early warning is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power customer service, and in particular to a big data mining method, system, device and medium for electric power service hot event analysis and prediction. BACKGROUND

[0002] Big data and artificial intelligence technology are becoming key means for improving customer satisfaction and preventing service risks in the field of electric power service. By integrating customer service work orders, online customer service, marketing systems and other multi-source data, a service risk control system can be built to realize real-time monitoring and analysis of customer demands. Currently, the industry has generally adopted methods such as rule-based sensitive word matching and work order quantity statistics for risk early warning and hierarchical control, preliminarily realizing the transition from passive response to active prevention and effectively supporting the service risk prevention and control of key business areas such as "new industrial workers" and "business environment".

[0003] However, the technical limitations of such processing methods are: single risk perception dimension: existing technologies mainly rely on text keywords for risk identification, which cannot capture the real emotional state of customers in the call, and there is a missed report for potential high-risk events with emotional agitation but without using sensitive words; weak risk prediction ability: existing methods are mostly static and isolated event analysis, lacking dynamic prediction ability for potential diffusion paths and evolution trends of risk events in customer groups, geographical areas and time dimensions, resulting in delayed early warning; rigid risk assessment mechanism: risk determination is usually based on fixed thresholds and rules, lacking a comprehensive assessment model that can quantitatively integrate multi-dimensional risk factors such as "individual emotional intensity" and "group diffusion breadth", resulting in insufficient accuracy of early warning and matching of management levels. SUMMARY

[0004] In view of the above problems, the present application is proposed.

[0005] To solve the above technical problems, the present application provides the following technical solutions: a big data mining method for electric power service hot event analysis and prediction, comprising: obtaining customer service data of a plurality of electric power service interaction systems, and constructing an electric power service risk comprehensive data set containing customer demands and service behaviors;

[0006] Based on the electric power service risk comprehensive data set, the joint analysis of voiceprint features and text semantics is performed on the customer interaction content, and the customer emotional score of each service event is generated;

[0007] Based on the electric power service risk comprehensive data set, a dynamic heterogeneous graph is constructed, and a graph neural network model is used to analyze the propagation path of risk events in the time and space dimensions, and a risk diffusion prediction result of each service event is generated;

[0008] According to the customer sentiment score and the risk diffusion prediction result, a comprehensive risk index of each service event is calculated;

[0009] When the comprehensive risk index exceeds a preset multi-level threshold value, a warning work order of a corresponding level is generated, and a message notification to personnel of a corresponding level is triggered according to a risk area and an event type indicated by the comprehensive risk index.

[0010] As a preferred scheme of the power service hotspot event analysis and prediction big data mining method, the constructing the power service risk comprehensive data set containing customer appeal and service behavior comprises:

[0011] Accessing raw customer service interaction data from customer service hotlines, online customer service platforms and service management systems to form a multi-source customer service data set;

[0012] Performing data cleaning and format normalization processing on the multi-source customer service data set to eliminate invalid and duplicate data, and generating a standardized customer service data set;

[0013] Extracting customer appeal text, service interaction time, service area identifier and service event type from the standardized customer service data set to construct the power service risk comprehensive data set containing customer appeal and service behavior.

[0014] As a preferred scheme of the power service hotspot event analysis and prediction big data mining method, the generating a customer sentiment score of each service event comprises:

[0015] Extracting an acoustic feature vector from the call recording data of the power service risk comprehensive data set, the acoustic feature vector comprising fundamental frequency, energy, speech rate and mel-frequency cepstral coefficient;

[0016] Extracting a text semantic feature vector from the customer appeal text data of the power service risk comprehensive data set, the text semantic feature vector being obtained by encoding through a pre-trained language model;

[0017] Inputting the acoustic feature vector and the text semantic feature vector into a multi-modal fusion neural network, calculating the weight of each modal feature through an attention mechanism, and generating a fusion feature vector;

[0018] Inputting the fusion feature vector into a fully connected classification layer, and outputting a value between 0 and 1 as the customer sentiment score of each service event.

[0019] As a preferred scheme of the power service hotspot event analysis and prediction big data mining method, the generating a risk diffusion prediction result of each service event comprises:

[0020] extracting customer identification, service area identification, service event type and service interaction time from the power service risk comprehensive dataset as node attributes for constructing a dynamic heterogeneous graph;

[0021] According to the repeated complaints of customers in the same service area, the concentrated occurrence of the same service event type, and the time continuity of service events in adjacent service areas, edges are constructed between customer nodes, service area nodes and service event nodes in the dynamic heterogeneous graph;

[0022] The service interaction time is used as the timestamp of the graph, and the dynamic heterogeneous graph is time-sliced to generate a series of discrete time snapshots;

[0023] Each time snapshot is input into a time series graph neural network model, the information of adjacent nodes is aggregated, the risk state vector of each service area node is calculated, and the risk diffusion prediction result of each service area in the next time window is output.

[0024] As a preferred scheme of the power service hotspot event analysis and prediction big data mining method, the edges between customer nodes, service area nodes and service event nodes are constructed, including:

[0025] According to the service interaction time of the power service risk comprehensive dataset, multiple service requests initiated by the same customer to the same service area within a preset time window are identified, and a "repeat complaint" edge is established between the customer node and the service area node;

[0026] The number of work orders of the service event type of the preset high-risk type in each service area node within a preset statistical period is counted, and when the number of work orders exceeds a preset threshold, a "concentrated occurrence" edge is established between the service area node and the corresponding service event node;

[0027] The occurrence time sequence of work orders of the same service event type between adjacent service area nodes is analyzed, and when the work order occurrence time of the first service area node is earlier than that of the second service area node and the time difference is within a preset diffusion delay range, a "spatiotemporal continuity" edge is established between the first service area node and the second service area node.

[0028] As a preferred scheme of the power service hotspot event analysis and prediction big data mining method, the calculation of the comprehensive risk index of each service event includes:

[0029] Obtain the customer sentiment score and risk diffusion prediction result of each service event, wherein the risk diffusion prediction result includes a set of predicted affected service area nodes and the risk level of each node;

[0030] According to the preset weight coefficient, the customer emotion score is linearly weighted to obtain an emotion risk component;

[0031] The number of nodes in a high risk level in the predicted influence service area node set is counted, and a regional diffusion risk component is obtained according to the product of the node number and a preset regional influence coefficient;

[0032] The emotion risk component and the regional diffusion risk component are added to calculate a comprehensive risk index of each service event.

[0033] As a preferred scheme of the big data mining method for analyzing and predicting the power service hotspot event, the generating a corresponding level warning work order comprises:

[0034] The comprehensive risk index is compared with a plurality of preset threshold values, when the comprehensive risk index is greater than or equal to a first threshold value, it is determined as a first level risk event; when the comprehensive risk index is less than the first threshold value and greater than or equal to a second threshold value, it is determined as a second level risk event; when the comprehensive risk index is less than the second threshold value and greater than or equal to a third threshold value, it is determined as a third level risk event; when the comprehensive risk index is less than the third threshold value and greater than or equal to a fourth threshold value, it is determined as a fourth level risk event;

[0035] According to the determination result of the first level risk event or the second level risk event, a provincial warning work order is generated; according to the determination result of the third level risk event or the fourth level risk event, a city warning work order is generated;

[0036] According to the provincial warning work order, an eLink message sending service is used to send a warning notification to a provincial dispatcher; according to the city warning work order, a short message sending service is used to send a warning notification to a city dispatcher.

[0037] The application provides a big data mining system for analyzing and predicting a power service hotspot event.

[0038] To solve the above technical problems, the application further provides the following technical scheme: a big data mining system for analyzing and predicting a power service hotspot event, comprising: a multi-source service data fusion module, configured to acquire customer service data of a plurality of power service interaction systems, and to construct a power service risk comprehensive data set containing customer demands and service behaviors;

[0039] A multi-modal emotion recognition module is configured to perform joint analysis of voiceprint features and text semantics based on the power service risk comprehensive data set, to generate a customer emotion score of each service event;

[0040] A graph risk analysis module is configured to construct a dynamic heterogeneous graph based on the power service risk comprehensive dataset, and analyze the propagation path of a risk event in a time-space dimension by using a graph neural network model to generate a risk diffusion prediction result of each service event;

[0041] A comprehensive risk assessment module is configured to calculate a comprehensive risk index of each service event according to the customer emotion score and the risk diffusion prediction result;

[0042] An intelligent early warning work order module is configured to generate an early warning work order of a corresponding level when the comprehensive risk index exceeds a preset multi-level threshold, and trigger a message notification to a corresponding level of dispatch personnel according to a risk area and an event type indicated by the comprehensive risk index.

[0043] The application provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the big data mining method for power service hotspot event analysis and prediction.

[0044] The application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the big data mining method for power service hotspot event analysis and prediction.

[0045] The application has the following beneficial effects: by constructing a power service risk comprehensive dataset containing customer demands and service behaviors, the fusion and structuring of multi-source heterogeneous service data are realized, and a comprehensive and accurate data foundation is laid for subsequent intelligent analysis; by jointly analyzing the voiceprint features and text semantics of customer interaction content based on the comprehensive dataset, the accurate quantitative evaluation of customer real emotions is realized, the limitations of traditional text analysis are broken through, and the depth and sensitivity of risk perception are significantly improved; by constructing a dynamic heterogeneous graph based on the comprehensive dataset and analyzing the propagation path of a risk event in a time-space dimension by using a graph neural network model, the dynamic prediction of the future influence range and diffusion trend of a risk event is realized, and the risk control is promoted from point-shaped early warning to planar prediction; by calculating a comprehensive risk index of each service event according to the customer emotion score and the risk diffusion prediction result, the multi-dimensional and quantifiable comprehensive evaluation of a risk event is realized, and an objective and comprehensive risk decision basis is provided; by generating an early warning work order of a corresponding level when the comprehensive risk index exceeds a preset multi-level threshold and triggering a message notification to a corresponding level of dispatch personnel, the automation, precision and hierarchical response of risk early warning are realized, the last mile from risk prediction to emergency disposal is broken through, and the beneficial effects of improving the timeliness, accuracy and management level matching degree of power service risk early warning are finally achieved. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0047] Figure 1 The overall flowchart of the big data mining method for power service hotspot event analysis and prediction provided by an embodiment of the present application is shown in FIG. 1.

[0048] Figure 2 The computer device diagram of the big data mining method for power service hotspot event analysis and prediction provided by an embodiment of the present application is shown in FIG. 2. DETAILED DESCRIPTION

[0049] In order to make the above-mentioned objects, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings in the specification. Obviously, the described embodiments are only some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.

[0050] Embodiment 1, refer to Figure 1 The first embodiment of the present application provides a big data mining method for power service hotspot event analysis and prediction, which comprises the following steps:

[0051] S1: Obtain customer service data of multiple power service interaction systems, and construct a power service risk comprehensive data set containing customer demands and service behaviors;

[0052] S2: Based on the power service risk comprehensive data set, perform joint analysis of voiceprint features and text semantics on customer interaction content, and generate a customer emotion score of each service event;

[0053] S3: Based on the power service risk comprehensive data set, construct a dynamic heterogeneous graph, and use a graph neural network model to analyze the propagation path of risk events in the time-space dimension, and generate a risk diffusion prediction result of each service event;

[0054] S4: Calculate a comprehensive risk index of each service event according to the customer emotion score and the risk diffusion prediction result;

[0055] S5: when the comprehensive risk index exceeds a preset multi-level threshold, a warning work order of a corresponding level is generated, and a message notification to dispatch personnel of a corresponding level is triggered according to a risk area and an event type indicated by the comprehensive risk index.

[0056] It should be noted that the traditional power service risk analysis relies on artificial experience or simple keyword matching rules, and the identification of customer complaints stays on the surface text, which is difficult to capture the real and potential emotional state of the customer and cannot predict the diffusion trend of the risk event, resulting in late warning and narrow coverage. At the same time, the static sensitive word library and the fixed risk threshold are difficult to adapt to the evolving customer expression and the dynamically changing external environment (such as weather, major events), so that the system has a blind area when dealing with new and complex risks, and the accuracy and timeliness of risk identification are insufficient.

[0057] Therefore, in order to solve the problems of shallow risk perception, lack of evolution prediction and poor system adaptability, through the steps of S1 to S5, a multi-modal fusion risk perception model is constructed, the precise quantification of customer emotional intensity is realized by joint analysis of voiceprint features and text semantics; a dynamic heterogeneous graph is constructed and a graph neural network is used to realize the prediction of the propagation path of the risk event in the time and space dimensions; the comprehensive risk index is calculated by weighted fusion of customer emotional score and risk diffusion prediction result, realizing the leap from single-dimensional judgment to multi-dimensional dynamic evaluation, and triggering a hierarchical warning based on the index, thereby realizing earlier, more accurate and more comprehensive analysis and prediction of power service hot events.

[0058] Embodiment 2, refer to Figure 1 For the second embodiment of the application, a big data mining method for power service hot event analysis and prediction is provided.

[0059] S1: obtaining customer service data of a plurality of power service interaction systems, and constructing a power service risk comprehensive data set containing customer complaints and service behaviors;

[0060] Specifically, S1 includes the following steps:

[0061] S11: accessing original customer service interaction data from a customer service system, a work order management system and an external public opinion platform, to form a multi-source customer service data set;

[0062] In an optional embodiment, the original customer service interaction data includes call recording, online customer service chat record, work order text description and social media text information.

[0063] In an optional embodiment, the accessing step pulls incremental data from the customer service system, the work order management system and the external public opinion platform at a minute level frequency by calling a preset data interface.

[0064] In an optional embodiment, the data interface adopts message queue middleware for data transmission to ensure real-time and reliability of data transmission.

[0065] S12: data cleaning and format normalization processing is performed on the multi-source customer service data set to eliminate invalid and duplicate data, and a standardized customer service data set is generated;

[0066] In an optional embodiment, the data cleaning includes speech recognition conversion to text for call recording, and filtering of special symbols and meaningless filler words in the text.

[0067] In an optional embodiment, the format normalization processing includes unifying date and time formats and region encoding formats from different systems into preset standard formats.

[0068] In an optional embodiment, the elimination of invalid and duplicate data includes data deduplication based on customer unique identifier and interaction timestamp, and filtering out invalid call records with call duration less than a preset threshold.

[0069] S13: customer complaint text, service interaction time, service area identifier and service event type are extracted from the standardized customer service data set to build a power service risk comprehensive data set containing customer complaints and service behaviors.

[0070] In an optional embodiment, the service event type includes service request, service complaint, service suggestion and service report.

[0071] In an optional embodiment, the service area identifier is obtained by matching the geographic coordinates of the customer service request with a preset power supply grid code, wherein the power supply grid code is pre-divided according to the topological relationship of transformer substations, lines and transformer areas.

[0072] In an optional embodiment, the customer complaint text, service interaction time, service area identifier and service event type are stored as data fields in a preset structured data table, and the primary key of the structured data table is the combination of service interaction time and customer unique identifier, to form the power service risk comprehensive data set.

[0073] S2: based on the power service risk comprehensive data set, joint analysis of voiceprint features and text semantics is performed on customer interaction content to generate customer emotion score for each service event;

[0074] Specifically, S2 includes the following steps:

[0075] S21: extracting an acoustic feature vector from the call recording data of the power service risk comprehensive dataset, the acoustic feature vector comprising a fundamental frequency, an energy, a speech rate, and a mel-frequency cepstral coefficient;

[0076] In an optional embodiment, the step of extracting the acoustic feature vector comprises frame segmentation and windowing processing on the call recording data.

[0077] In an optional embodiment, the frame length of the frame segmentation is 25 milliseconds, the frame shift is 10 milliseconds, and the Hamming window is used for windowing processing.

[0078] In an optional embodiment, based on the windowed voice frame sequence, the fundamental frequency of each frame is calculated by the autocorrelation function method, the energy is calculated by the sum of squares method, the speech rate is calculated by the short-time zero-crossing rate, and the mel-frequency cepstral coefficient is obtained by the mel filter bank and the discrete cosine transform.

[0079] S22: extracting a text semantic feature vector from the customer complaint text data of the power service risk comprehensive dataset, the text semantic feature vector being obtained by encoding through a pre-trained language model;

[0080] In an optional embodiment, the pre-trained language model is a bidirectional encoder representation model based on a Transformer architecture.

[0081] In an optional embodiment, the customer complaint text data is input into the bidirectional encoder representation model to obtain a hidden state vector of all word pieces in the last layer of the model.

[0082] In an optional embodiment, the hidden state vector is subjected to mean pooling processing to generate a fixed-dimension text semantic feature vector.

[0083] S23: inputting the acoustic feature vector and the text semantic feature vector into a multi-modal fusion neural network, calculating the weight of each modal feature through an attention mechanism, and generating a fusion feature vector;

[0084] In an optional embodiment, the multi-modal fusion neural network comprises a cross-attention layer for calculating the attention weight of the acoustic feature vector on the text semantic feature vector.

[0085] In an optional embodiment, the cross-attention layer generates a weighted text semantic feature through the calculation mode of query, key, and value, taking the acoustic feature vector as the query, the text semantic feature vector as the key and the value.

[0086] In an optional embodiment, the weighted text semantic feature and the original acoustic feature vector are spliced to input a fully connected layer for nonlinear transformation, and a fusion feature vector is output.

[0087] S24: inputting the fusion feature vector into a fully connected classification layer to output a value between 0 and 1 as a customer sentiment score of each service event.

[0088] In an optional embodiment, the fully connected classification layer comprises a hidden layer and an output layer, the hidden layer adopts a ReLU activation function, and the output layer adopts a Sigmoid activation function.

[0089] In an optional embodiment, the input of the Sigmoid activation function is the output value of the fusion feature vector after being transformed by the hidden layer.

[0090] In an optional embodiment, the output value of the Sigmoid activation function is taken as the customer sentiment score, and the closer the score is to 1, the more the customer sentiment tends to be negative.

[0091] S3: based on the power service risk comprehensive dataset, constructing a dynamic heterogeneous graph, and using a graph neural network model to analyze the propagation path of the risk event in the time and space dimensions to generate a risk diffusion prediction result of each service event;

[0092] Specifically, S3 comprises the following steps:

[0093] S31: extracting customer identification, service area identification, service event type, and service interaction time from the power service risk comprehensive dataset as node attributes for constructing the dynamic heterogeneous graph;

[0094] In an optional embodiment, the customer identification is a customer unique code generated by the system, the service area identification is a power supply grid code based on power grid topology division, the service event type is a preset work order classification code, and the service interaction time is a standardized timestamp.

[0095] In an optional embodiment, the customer identification, service area identification, and service event type are taken as attributes of customer nodes, service area nodes, and service event nodes respectively, and the service interaction time is taken as an attribute of edges associated with these nodes.

[0096] In an optional embodiment, the service interaction time is discretized and divided into time windows in units of hours for subsequent graph time slicing.

[0097] S32: constructing edges between customer nodes, service area nodes, and service event nodes in the dynamic heterogeneous graph according to repeated complaints of customers in the same service area, concentrated occurrence of the same service event type, and time continuity of service events in adjacent service areas;

[0098] Specifically, S32 further comprises the following steps:

[0099] S32a: According to the service interaction time of the power service risk comprehensive data set, identifying multiple service requests initiated by the same customer to the same service area within a preset time window, and establishing a "repeat appeal" edge between the customer node and the service area node;

[0100] In an optional embodiment, the preset time window is 24 hours.

[0101] In an optional embodiment, when the number of service requests initiated by the same customer identifier to the same service area identifier within 24 hours is greater than or equal to 3 times, it is determined as a repeat appeal.

[0102] In an optional embodiment, the weight of the "repeat appeal" edge is set to the cumulative number of service requests.

[0103] S32b: Count the number of work orders of a preset high-risk type of service event in a preset statistical period for each service area node, and when the number of work orders exceeds a preset threshold, establish a "concentrated occurrence" edge between the service area node and the corresponding service event node;

[0104] In an optional embodiment, the preset high-risk type includes service complaints and service reports.

[0105] In an optional embodiment, the preset statistical period is 7 days, and the preset threshold is 5.

[0106] In an optional embodiment, the weight of the "concentrated occurrence" edge is set to the number of work orders exceeding the preset threshold.

[0107] S32c: Analyze the occurrence time sequence of work orders of the same service event type between adjacent service area nodes, and when the occurrence time of the work order of the first service area node is earlier than that of the second service area node and the time difference is within a preset diffusion delay range, establish a "spatiotemporal continuity" edge between the first service area node and the second service area node.

[0108] In an optional embodiment, the geographical adjacency relationship between two service area nodes is determined by judging whether the power grid codes of the two service area nodes belong to adjacent lines under the same transformer substation.

[0109] In an optional embodiment, the preset diffusion delay range is 1 hour to 48 hours.

[0110] In an optional embodiment, the weight of the "spatiotemporal continuity" edge is set to the inverse of the time difference between the two service area nodes.

[0111] S33: Time-slice the dynamic heterogeneous graph atlas by taking the service interaction time as the timestamp of the atlas, to generate a series of discrete time snapshots;

[0112] In an optional embodiment, the granularity of the time-slicing is 1 hour.

[0113] In an optional embodiment, for each time window, all nodes and edges falling within the window form an independent graph structure, which is taken as a snapshot of the graph atlas at that time.

[0114] In an optional embodiment, the sequence of N consecutive time snapshots is input into a time-series graph neural network model for processing.

[0115] S34: Input each time snapshot into a time-series graph neural network model, aggregate the information of neighboring nodes, calculate the risk state vector of each service area node, and output the risk diffusion prediction result of each service area in the next time window.

[0116] In an optional embodiment, the time-series graph neural network model is a combination model of a graph convolution network and a gated recurrent unit.

[0117] In an optional embodiment, the aggregation of the information of neighboring nodes includes weighted summation of the risk states of the neighboring nodes connected by the "repeated appeal" edge, the "concentrated occurrence" edge, and the "spatiotemporal continuity" edge.

[0118] In an optional embodiment, the risk diffusion prediction result is a vector representing the risk level of each service area node in the next time window, and the risk level includes low risk, medium risk, and high risk.

[0119] S4: Calculate the comprehensive risk index of each service event according to the customer sentiment score and the risk diffusion prediction result;

[0120] Specifically, S4 includes the following steps:

[0121] S41: Obtain the customer sentiment score and the risk diffusion prediction result of each service event, wherein the risk diffusion prediction result includes a set of predicted-affected service area nodes and the risk level of each node.

[0122] In an optional embodiment, the customer sentiment score ranges from 0 to 1, and the risk level includes low risk, medium risk, and high risk.

[0123] In an optional embodiment, when the risk level is high risk, the corresponding service area node is added to the set of predicted-affected service area nodes.

[0124] In an optional embodiment, the risk diffusion prediction result further comprises a risk state vector of each service area node.

[0125] S42: Linearly weighting the customer sentiment score according to a preset weight coefficient to obtain an emotional risk component;

[0126] In an optional embodiment, the preset weight coefficient is 0.6.

[0127] In an optional embodiment, the calculation formula of the linear weighting is: emotional risk component = customer sentiment score x weight coefficient.

[0128] In an optional embodiment, the weight coefficient is determined by a grid search algorithm according to historical early warning accuracy.

[0129] S43: Counting the number of nodes in the predicted affected service area node set that are in a high risk level, and obtaining a regional diffusion risk component according to the product of the number of nodes and a preset regional influence coefficient.

[0130] In an optional embodiment, the preset regional influence coefficient is 1.0.

[0131] In an optional embodiment, the calculation formula of the regional diffusion risk component is: regional diffusion risk component = number of nodes in a high risk level x regional influence coefficient.

[0132] In an optional embodiment, the regional influence coefficient is weighted and set according to the number of customers of the service area node, and the more the number of customers, the greater the regional influence coefficient.

[0133] S44: Adding the emotional risk component and the regional diffusion risk component to calculate a comprehensive risk index of each service event.

[0134] In an optional embodiment, the calculation formula of the comprehensive risk index is: comprehensive risk index = emotional risk component + regional diffusion risk component.

[0135] In an optional embodiment, the comprehensive risk index is a non-negative real number.

[0136] In an optional embodiment, when the comprehensive risk index is greater than 2.0, it is determined as a major risk event.

[0137] S5: When the comprehensive risk index exceeds a preset multi-level threshold, a corresponding level of early warning work order is generated, and a message notification to the corresponding level of dispatch personnel is triggered according to the risk area and event type indicated by the comprehensive risk index.

[0138] Specifically, S5 comprises the following steps:

[0139] S51: comparing the comprehensive risk index with preset multi-level thresholds, determining a first-level risk event when the comprehensive risk index is greater than or equal to a first threshold, determining a second-level risk event when the comprehensive risk index is less than the first threshold and greater than or equal to a second threshold, determining a third-level risk event when the comprehensive risk index is less than the second threshold and greater than or equal to a third threshold, and determining a fourth-level risk event when the comprehensive risk index is less than the third threshold and greater than or equal to a fourth threshold;

[0140] In an optional embodiment, the first threshold is 4.0, the second threshold is 3.0, the third threshold is 2.0, and the fourth threshold is 1.0.

[0141] In an optional embodiment, when the comprehensive risk index is greater than or equal to 4.0, a first-level risk event is determined, indicating that the risk has a wide range of influence and a high degree of urgency.

[0142] In an optional embodiment, when the comprehensive risk index is less than 4.0 and greater than or equal to 3.0, a second-level risk event is determined, indicating that the risk has a wide range of influence but a moderate degree of urgency.

[0143] S52: generating a provincial warning work order according to the determination result of the first-level risk event or the second-level risk event, and generating a prefectural warning work order according to the determination result of the third-level risk event or the fourth-level risk event.

[0144] In an optional embodiment, the provincial warning work order comprises event details, a risk area, an event type, a comprehensive risk index, and suggested disposal measures.

[0145] In an optional embodiment, the prefectural warning work order comprises event details, a risk area, an event type, and a comprehensive risk index.

[0146] In an optional embodiment, the step of generating a warning work order comprises calling a preset work order template and filling risk event information into the template.

[0147] S53: sending a warning notification to provincial dispatch personnel through an eLink message sending service according to the provincial warning work order, and sending a warning notification to prefectural dispatch personnel through a short message sending service according to the prefectural warning work order.

[0148] In an optional embodiment, the eLink message sending service receives a read receipt of the provincial dispatch personnel after sending the warning notification.

[0149] In an optional embodiment, the short message sending service calls a preset short message template, embeds the key information of the early warning work order into the template, and sends the template.

[0150] In an optional embodiment, when no read receipt is received within a preset time or the short message sending fails, an automatic outbound service is started to send a voice early warning notice to dispatch personnel of a corresponding level.

[0151] Embodiment 3 is a third embodiment of the present application, which provides a big data mining system for power service hotspot event analysis and prediction, comprising:

[0152] A multi-source service data fusion module is configured to acquire customer service data of multiple power service interaction systems, and construct a power service risk comprehensive data set containing customer demands and service behaviors;

[0153] A multi-modal emotion recognition module is configured to perform joint analysis of voiceprint features and text semantics on customer interaction content based on the power service risk comprehensive data set, and generate a customer emotion score of each service event;

[0154] A graph risk analysis module is configured to construct a dynamic heterogeneous graph based on the power service risk comprehensive data set, and analyze the propagation path of a risk event in a time-space dimension by using a graph neural network model, and generate a risk diffusion prediction result of each service event;

[0155] A comprehensive risk assessment module is configured to calculate a comprehensive risk index of each service event according to the customer emotion score and the risk diffusion prediction result;

[0156] An intelligent early warning work order module is configured to generate an early warning work order of a corresponding level when the comprehensive risk index exceeds a preset multi-level threshold, and trigger a message notice to dispatch personnel of a corresponding level according to a risk area and an event type indicated by the comprehensive risk index.

[0157] Embodiment 4, refer to Figure 2For the fourth embodiment of the present application, which is different from the first three embodiments, the function, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application, in essence or the part that contributes to the prior art, or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0158] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logic functions, and can be specifically embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other system that can take instructions from an instruction execution system, apparatus, or device, or in conjunction with these instructions execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0159] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electrical devices), a portable computer diskette (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or other suitable medium on which the program can be printed, as the program can be electronically obtained, for example, by optical scanning of the paper or other medium, followed by editing, interpreting, or otherwise processing, if necessary, in other suitable ways to be electronically obtained, and then stored in the computer memory.

[0160] It should be understood that portions of the present application can be implemented in hardware, software, firmware, or combinations thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, implementation can be with any or a combination of the following technologies, which are all well known in the art: a discrete logic circuit having logic gates for implementing logic functions upon an application of data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0161] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and all should be covered in the scope of the claims of the present application.

Claims

1. A big data mining method for analyzing and predicting hot events in power services, characterized in that: include, Acquire customer service data from multiple power service interaction systems to construct a comprehensive power service risk dataset that includes customer requests and service behaviors; Based on the aforementioned comprehensive dataset of power service risks, a joint analysis of voiceprint features and text semantics is performed on customer interaction content to generate a customer sentiment score for each service event. Based on the aforementioned comprehensive dataset of power service risks, a dynamic heterogeneous graph is constructed, and a graph neural network model is used to analyze the propagation path of risk events in the spatiotemporal dimension, generating risk diffusion prediction results for each service event. Based on the customer sentiment score and the risk diffusion prediction results, calculate the comprehensive risk index for each service event; When the comprehensive risk index exceeds the preset multi-level threshold, a warning work order of the corresponding level is generated, and a message notification is triggered to the corresponding level of dispatcher based on the risk area and event type indicated by the comprehensive risk index.

2. The big data mining method for analyzing and predicting hot events in power services as described in claim 1, characterized in that: The construction of the comprehensive dataset of power service risks, which includes customer demands and service behaviors, includes: Access raw customer service interaction data from customer service hotlines, online customer service platforms, and service management systems to form a multi-source customer service dataset; The multi-source customer service dataset is cleaned and formatted to remove invalid and duplicate data, generating a standardized customer service dataset. The standardized customer service dataset is used to extract customer request text, service interaction time, service area identifier and service event type to construct a comprehensive power service risk dataset that includes customer requests and service behaviors.

3. The big data mining method for analyzing and predicting hot events in power services as described in claim 2, characterized in that: The customer sentiment score generated for each service event includes: Acoustic feature vectors are extracted from call recording data in the comprehensive dataset of power service risks. The acoustic feature vectors include fundamental frequency, energy, speech rate, and Mel frequency cepstral coefficients. Text semantic feature vectors are extracted from customer request text data in the comprehensive dataset of power service risks. These text semantic feature vectors are obtained by encoding through a pre-trained language model. The acoustic feature vector and the text semantic feature vector are input into a multimodal fusion neural network, and the weights of each modality feature are calculated through an attention mechanism to generate a fused feature vector. The fused feature vector is input into a fully connected classification layer, and a value between 0 and 1 is output as the customer sentiment score for each service event.

4. The big data mining method for analyzing and predicting hot events in power services as described in claim 3, characterized in that: The risk diffusion prediction results generated for each service event include: Customer identifiers, service area identifiers, service event types, and service interaction times are extracted from the comprehensive dataset of power service risks and used as node attributes for constructing a dynamic heterogeneous graph. Based on repeated customer requests in the same service area, the concentrated occurrence of the same type of service event, and the temporal continuity of service events in adjacent service areas, edges are constructed between customer nodes, service area nodes, and service event nodes in the dynamic heterogeneous graph. Using the service interaction time as the timestamp of the graph, the dynamic heterogeneous graph is sliced ​​over time to generate a series of discrete time snapshots. Each time snapshot is input into the time-series graph neural network model. By aggregating the information of neighboring nodes, the risk state vector of each service area node is calculated, and the risk diffusion prediction results of each service area in the next time window are output.

5. The big data mining method for analyzing and predicting hot events in power services as described in claim 4, characterized in that: The edges connecting the client node, service area node, and service event node include: Based on the service interaction time of the power service risk comprehensive dataset, identify multiple service requests initiated by the same customer to the same service area within a preset time window, and establish a "repeated request" edge between the customer node and the service area node. The number of work orders of the service event type of the preset high-risk type is counted for each service area node within a preset statistical period. When the number of work orders exceeds a preset threshold, a "centralized occurrence" edge is established between the service area node and the corresponding service event node. Analyze the occurrence time sequence of work orders of the same service event type between adjacent service area nodes. When the work order of the first service area node occurs earlier than that of the second service area node and the time difference is within the preset diffusion delay range, establish a "spatiotemporal continuity" edge between the first service area node and the second service area node.

6. The big data mining method for analyzing and predicting hot events in power services as described in claim 5, characterized in that: The calculation of the comprehensive risk index for each service event includes: Obtain customer sentiment scores and risk diffusion prediction results for each service event, wherein the risk diffusion prediction results include a set of service area nodes predicted to be affected and the risk level of each node; The customer's emotional score is linearly weighted according to a preset weighting coefficient to obtain an emotional risk component. The number of nodes at the high-risk level in the service area node set of the predicted impact is counted, and the regional diffusion risk component is obtained by multiplying the number of nodes by the preset regional impact coefficient. The emotional risk component is added to the regional diffusion risk component to calculate the comprehensive risk index for each service event.

7. The big data mining method for analyzing and predicting hot events in power services as described in claim 6, characterized in that: The generation of the corresponding level of early warning work order includes: The comprehensive risk index is compared with preset multi-level thresholds. When the comprehensive risk index is greater than or equal to the first threshold, it is determined to be a level 1 risk event; when the comprehensive risk index is less than the first threshold but greater than or equal to the second threshold, it is determined to be a level 2 risk event; when the comprehensive risk index is less than the second threshold but greater than or equal to the third threshold, it is determined to be a level 3 risk event; and when the comprehensive risk index is less than the third threshold but greater than or equal to the fourth threshold, it is determined to be a level 4 risk event. Based on the determination results of the Level 1 or Level 2 risk events, a provincial-level early warning work order is generated; based on the determination results of the Level 3 or Level 4 risk events, a municipal-level early warning work order is generated. Based on the provincial-level early warning work order, an early warning notification is sent to the provincial-level dispatcher via the eLink message sending service; based on the municipal-level early warning work order, an early warning notification is sent to the municipal-level dispatcher via the SMS sending service.

8. A big data mining system for analyzing and predicting hot events in power services, employing the big data mining method for analyzing and predicting hot events in power services as described in any one of claims 1 to 7, characterized in that, include: The multi-source service data fusion module is used to acquire customer service data from multiple power service interaction systems and construct a comprehensive power service risk dataset that includes customer demands and service behaviors. The multimodal emotion recognition module is used to perform joint analysis of voiceprint features and text semantics on customer interaction content based on the comprehensive dataset of power service risks, and generate a customer emotion score for each service event. The graph risk analysis module is used to construct a dynamic heterogeneous graph based on the comprehensive dataset of power service risks, and to use a graph neural network model to analyze the propagation path of risk events in the spatiotemporal dimension, generating risk diffusion prediction results for each service event. The comprehensive risk assessment module is used to calculate the comprehensive risk index for each service event based on the customer sentiment score and the risk diffusion prediction results. The intelligent early warning work order module is used to generate an early warning work order of the corresponding level when the comprehensive risk index exceeds a preset multi-level threshold, and to trigger a message notification to the corresponding level of dispatchers based on the risk area and event type indicated by the comprehensive risk index.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the big data mining method for analyzing and predicting hot events in power services as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the big data mining method for analyzing and predicting hot events in power services as described in any one of claims 1 to 7.

Images