Intelligent pushing method and device for electric power service information and computer program product

Abstract

The invention discloses an electric power service information intelligent pushing method and device and a computer program product, and the method comprises the steps: S1, collecting electric power service related multi-source heterogeneous data; s2, constructing an electric power service knowledge graph based on the multi-source heterogeneous data, wherein the knowledge graph comprises a user entity, an equipment entity, a service entity, a power grid node entity and relationships and dynamic weight parameters among the user entity, the equipment entity, the service entity and the power grid node entity; s3, generating a user portrait based on the electric power service knowledge graph, wherein the user portrait comprises power utilization characteristics, service demand preferences, user type labels and credit levels; s4, on the basis of the user portrait and the power grid operation state, demand identification and scene matching are carried out through the power service knowledge graph to determine a push strategy; and S5, pushing the power service information to the user terminal according to the pushing strategy. The technical problems of poor accuracy, passive response, lack of association and the like existing in traditional power service information pushing are effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power systems, in particular to a power service information intelligent pushing method and device and computer program product. BACKGROUND

[0002] With the rapid development of emerging technologies such as big data, cloud computing, smart grid and energy internet, the power system has generated a large amount of multi-source and heterogeneous operation and service data covering multiple dimensions such as equipment operation status, user power consumption behavior, power grid topology structure, policy and regulation text and historical work order records. The complexity and scale of these data provide a solid foundation for the intelligentization of power services, but also put forward higher requirements for information pushing technology.

[0003] The traditional power service information pushing mode (such as mass short message sending and bulletin board posting) has significant technical limitations: first, there is a lack of accurate analysis mechanism based on user portrait, and the broadcast pushing strategy leads to information overload and user resistance; second, the service mode is mainly passive response, and only provides information after the user initiates consultation, which is seriously insufficient in timeliness and foresight; third, the pushing content lacks multi-dimensional data correlation capability, and cannot effectively integrate information such as user historical behavior, equipment status and geographic location to form personalized solutions; fourth, the pushing decision-making process lacks explainability, and users cannot understand the logical basis of information pushing; finally, the information pushing channels are scattered and fragmented, and a unified intelligent pushing platform has not been formed, resulting in fragmented service experience.

[0004] Although the rule engine or simple user tag system used by some current power enterprises has realized basic information distribution, its pushing logic is rigid and cannot adapt to the dynamic changes of power service scenarios, and it cannot mine the deep semantic association between data. This kind of solution cannot effectively handle complex scenarios such as multi-condition combination, time sequence dependence and context association in power service, and cannot meet the demand for precision and intelligence of modern power service.

[0005] At the same time, the demand and expectation of power consumers for information pushing mainly focus on timeliness, accuracy, personalization, convenience and interactivity, and the service experience demand presents the characteristics of differentiation, interaction and intelligence, and the customer service channel is no longer limited to traditional channels such as call hotline and business hall, but develops towards a more diversified and intelligent trend. The power consumers' demand for personalized and precise power service is evolving. There is an obvious contradiction between the demand of power consumers and the current situation of power service information pushing. SUMMARY

[0006] The technical problem to be solved by the embodiments of the present application is to provide a power service information intelligent pushing method, device and computer program product to realize personalized and precise pushing of power service information.

[0007] To solve the above technical problems, the present application provides an intelligent power service information pushing method, comprising: Step S1, collecting power service related multi-source heterogeneous data; Step S2, constructing a power service knowledge graph based on multi-source heterogeneous data, the knowledge graph including user entities, device entities, service entities, grid node entities, and their relationships and dynamic weight parameters; Step S3, generating a user portrait based on the power service knowledge graph, the user portrait including power consumption characteristics, service demand preferences, user type labels, and credit levels; Step S4, based on the user portrait and grid operation state, demand identification and scenario matching are performed through the power service knowledge graph to determine a pushing strategy; Step S5, pushing power service information to user terminals according to the pushing strategy.

[0008] Preferably, in step S4, the demand identification specifically includes: Defining core operation quantitative indicators of target devices, calculating indicator values based on device rated parameters and real-time operation data, setting multi-stage state threshold intervals, and starting associated reasoning when the indicator values reach preset trigger thresholds; Defining user and target device association degree parameters, determining association degree values based on entity relationship chains in the power service knowledge graph and cross-validation data, and screening user sets that meet the association degree threshold; Building a weighted calculation model containing device state quantitative normalized values, entity association degrees, and state duration normalized values, determining dimension weight coefficients, and calculating demand confidence, when the confidence reaches a set threshold, determining that the user has corresponding demand; Based on entity relationship traversal and rule engine reasoning of the power service knowledge graph, a traceable demand reasoning path is generated, the reasoning chain is bound to the user demand and stored.

[0009] Preferably, in the device state quantification, the calculation formula of the core operation quantitative indicator is: Indicator value = device real-time operation parameter / device rated parameter x 100% The device real-time operation parameter is obtained through a power Internet of Things collection terminal, and the state threshold interval includes at least three stages of normal state, warning state, and alarm state.

[0010] Preferably, the value rule of the entity correlation degree is: when the correlation degree = 1, it represents that the topology link of the user and the target device is completely matched; when the correlation degree = 0.5, it represents that the ownership relationship of the user and the target device is fuzzy matched; and when the correlation degree = 0, it represents that there is no explicit topology correlation; and the cross-validation data includes power GIS topology data and marketing profile data.

[0011] Preferably, the generation process of the inference chain includes: extracting the target device entity, the user entity and the correlation therebetween from the power service knowledge graph, and completing the initialization of the entity and the relationship; screening the device set in the abnormal state, marking the device state and the abnormal duration; traversing the user entity, tracing the associated target device, and screening the user set satisfying the correlation degree threshold; calling the rule engine, inputting the normalized value of the quantitative index, the correlation degree and the normalized value of the duration, and calculating the demand confidence; when the confidence reaches the judgment threshold, a standardized inference path is generated, which is bound to the user demand and stored in the knowledge graph demand inference entity attribute.

[0012] Preferably, in the step S4, the scene matching specifically includes: mapping the power grid operation state index, the meteorological condition parameter and the user demand confidence to the [0, 1] interval to generate a normalized feature set; presetting a power service scene rule, each rule containing a feature combination condition and a corresponding service scene; introducing a feature weight coefficient to calculate the scene matching degree; when the scene matching degree is greater than a preset scene triggering threshold, the corresponding service scene is matched; when multiple scenes are satisfied, the scene with the highest matching degree is selected as the output.

[0013] Preferably, the scene matching process further includes constructing a power grid-meteorological correlation rule, specifically: setting a coupling triggering threshold of the power grid operation index and the meteorological condition, the power grid operation index at least including the device load rate and the line voltage, and the meteorological condition at least including high temperature and heavy rain; introducing an associated influence coefficient to correct the scene matching degree, when the meteorological condition aggravates the power grid anomaly, the associated influence coefficient is greater than 1; when the meteorological condition has no influence, the associated influence coefficient is 1; if the power grid operation index satisfies the threshold but the meteorological condition does not satisfy, the associated influence coefficient is less than 1, and when the corrected scene matching degree satisfies a preset reverse verification threshold, the corresponding service scene is triggered.

[0014] Preferably, in the step S4, the push strategy comprises a push content priority, a push time window and a push channel, the push content priority is determined based on a user demand confidence and a power grid abnormality degree, the push time window is dynamically adjusted based on a user power consumption peak period and a power grid operation state, and the push channel is determined according to a push channel preference in the user portrait.

[0015] Preferably, in the step S4, the power grid operation data influences the push strategy from three dimensions of a push priority, push content and push timeliness, and specifically comprises: a power grid operation index-based abnormality degree is divided into levels, a higher abnormality degree corresponds to a higher push priority, and corresponding push time limit requirements are set for different abnormality levels; exclusive content is customized according to an abnormality type reflected by the power grid operation data; a push timing is dynamically adjusted in combination with real-time changes of the power grid operation data, the push is triggered immediately when the power grid operation index presents a continuous deterioration trend, and the push is delayed to a non-power consumption peak period when the index is stable in an early warning state and has no deterioration trend.

[0016] Preferably, the step S2 specifically comprises: feature attributes of user entities, device entities, service entities and power grid node entities are extracted; association relationships among the entities are determined; a dynamic weight parameter is assigned to each relationship, and the dynamic weight parameter is calculated based on a time decay factor, an association strength coefficient and a user feedback weight.

[0017] Preferably, the method further comprises a feedback optimization step, specifically comprising: interaction feedback data of the user on the push information is collected through the user terminal, the interaction feedback data comprises clicking, viewing, collecting, ignoring and feedback messages, wherein the user clicking a target service related link is regarded as valid feedback, and multiple ignorances of a certain type of push are regarded as no corresponding demand; newly added data and the interaction feedback data are regularly integrated into the power service knowledge graph to update entity attributes and relationships among the entities in the graph; a reinforcement learning algorithm is adopted, and user interaction conversion rate and user satisfaction score are used as reward functions to optimize push content screening rules, push timing judgment models and push channel selection strategies.

[0018] Preferably, the power service related multi-source heterogeneous data comprises user base data, power consumption behavior data, power grid operation data, power business data and external correlation data, wherein the power consumption behavior data comprises power consumption, power consumption period, power consumption mode and power consumption anomaly record, the power grid operation data comprises power grid topology structure, equipment operation state and fault record, the power business data comprises service request record, work order processing record and service evaluation record, and the external correlation data comprises policy and regulation text, meteorological data and social and economic data.

[0019] The application further provides a power service information intelligent pushing device, comprising: a collection module configured to collect power service related multi-source heterogeneous data; a construction module configured to construct a power service knowledge graph based on the multi-source heterogeneous data, wherein the knowledge graph comprises user entities, equipment entities, service entities, power grid node entities and relationships and dynamic weight parameters therebetween; a generation module configured to generate a user portrait based on the power service knowledge graph, wherein the user portrait comprises power consumption features, service demand preferences, user type labels and credit levels; a decision module configured to determine a pushing strategy by performing demand identification and scene matching based on the user portrait and power grid operation state through the power service knowledge graph; a pushing module configured to push power service information to a user terminal according to the pushing strategy.

[0020] The application further provides a power service information intelligent pushing device, comprising: one or more processors; a memory; one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, and the one or more application programs are configured to execute the power service information intelligent pushing method.

[0021] The application further provides a computer program product comprising computer instructions instructing a computer device to perform operations corresponding to the method.

[0022] The implementation of the present application has the following beneficial effects: the present application realizes the fundamental change of power service information push from passive response to active prediction by constructing an intelligent push strategy decision mechanism based on a power service knowledge graph. The mechanism innovatively combines user portraits and real-time operation states of the power grid to establish a dynamic decision-making closed loop of demand identification and scenario matching: firstly, the current power demand of the user is accurately captured, the potential demand is inferred in combination with the entity relationship in the knowledge graph, and the demand urgency and feasibility are evaluated in combination with real-time parameters such as power grid load rate, equipment availability and fault influence range; secondly, the matching degree of the user demand and the service entity is calculated, the demand urgency, service availability, user preference and power grid carrying capacity are comprehensively evaluated by a multi-dimensional weighted scoring model, and adaptive generation of the push strategy is realized; at the same time, the decision-making process is interpretable, and the system automatically labels the push basis to make the push logic transparent. The mechanism effectively solves the core problems of poor accuracy, information overload and lack of association in traditional push, significantly improves the pertinence and initiative of information push, changes the service from after-the-fact remedy to pre-emptive prevention, greatly reduces the demand for manual intervention, builds an intelligent and interpretable precise push core capability for power enterprises, and comprehensively supports the transformation of power services to active, personalized and intelligent. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. 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 labor.

[0024] Figure 1 is a flowchart of a power service information intelligent push method according to an embodiment of the present application.

[0025] Figure 2 is a flowchart of a push strategy decision according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following description of each embodiment is with reference to the drawings to illustrate specific embodiments in which the present application can be implemented.

[0027] Please refer to Figure 1 , an embodiment of the present application provides a power service information intelligent push method, which comprises: Step S1, collecting power service related multi-source heterogeneous data; Step S2, constructing a power service knowledge graph based on the multi-source heterogeneous data, wherein the knowledge graph comprises user entities, device entities, service entities, power grid node entities and the relationships and dynamic weight parameters therebetween; Step S3, generating a user portrait based on the power service knowledge graph, the user portrait including power consumption characteristics, service demand preferences, user type labels, and credit levels; Step S4, based on the user portrait and power grid operation state, demand identification and scene matching are performed through the power service knowledge graph to determine a push strategy; Step S5, power service information is pushed to a user terminal according to the push strategy.

[0028] Through the above steps, the power service information intelligent push method of the embodiment constructs a power global knowledge graph, actively and accurately pushes power consumption analysis, energy efficiency suggestions, fault warnings, emergency notifications, policy interpretations, and other information to users through real-time sensing and reasoning, and improves power service quality and user satisfaction. Among them, the power knowledge graph is like an intelligent relationship network that absorbs user power consumption, power grid state, and other data in real time, and through automatic adjustment of key weight parameters, the system can "sense" who should be given information and what information should be pushed.

[0029] The specific process of the power service information intelligent push method of the present application is described below.

[0030] Step S1 collects power service related multi-source heterogeneous data to provide data support for knowledge graph construction and push strategy formulation. The power service related multi-source heterogeneous data specifically includes: User basic data: collected through a power marketing system, including user ID, user type (residents / enterprises / parks), power consumption address, power consumption capacity, contact information, etc.

[0031] Power consumption behavior data: collected through smart meters and power consumption collection terminals, including real-time power consumption, power consumption time period distribution, load curve, payment records, power outage perception records, etc.

[0032] Power grid operation data: collected through power grid dispatching systems and distribution network automation systems, including transformer load rate, line current and voltage, transformer area power supply range, power grid fault location and type, etc.

[0033] Power business data: collected through a power business handling platform, including electricity payment, fault repair, power consumption installation, demand response participation records, photovoltaic grid connection application, etc.

[0034] External associated data: collected through a meteorological platform and a government policy release platform, including high temperature / rainstorm weather warnings, electricity price adjustment policies, demand response subsidy policies, regional industrial economic data, etc.

[0035] Step S2 constructs a power service knowledge graph.

[0036] First, define core entities, relationships, and set dynamic weight parameters.

[0037] The core entities of the power service knowledge graph include: user entities (attributes: user ID, power consumption type, power consumption capacity, historical fault record, service preference, etc.), device entities (attributes: device ID, device type, installation location, running time, rated parameter, real-time running state, etc.), service entities (attributes: service ID, service type, service content, applicable scenario, service timeliness, etc.), and power grid node entities (attributes: node ID, node type, load capacity, connection relationship, etc.).

[0038] The core relationships between entities include: user-device association relationships (such as "use", "belong to"), user-service association relationships (such as "have handled", "to be handled", "adapt to"), device-service association relationships (such as "need", "correspond to", "operation and maintenance"), device-power grid node association relationships (such as "access", "belong to"), and service-service association relationships (such as "association", "complement").

[0039] Each relationship corresponds to a weight parameter, and the weight value range is [0, 1], which is used to represent the closeness or importance of the relationship. The core weight parameters include: user demand intensity weight (representing the demand urgency of the user for a certain type of service), device risk level weight (representing the probability and impact of device failure), service adaptation degree weight (representing the matching degree of the service and the user / device), and relationship timeliness weight (representing the time decay coefficient of the association relationship).

[0040] The present application adopts the Flink streaming calculation framework to realize low-delay dynamic updating of entities, relationships and weights, and adapts to the real-time iteration needs of multi-source data in the power service scenario. The specific process is as follows: (1) Data preprocessing: using Flink Map / Filter operators to complete data cleaning (filtering null values, duplicate data), entity normalization (unifying entity identifiers such as user ID and device number), and relationship type marking (such as "user-repair-device", "user-subscription-service"). Standardize data of different formats into a unified format.

[0041] (2) Entity extraction: based on a model (such as BERT+CRF model), extract entities from preprocessed data, automatically create nodes and initialize attributes when new entities are detected; update graph node information directly when existing entity attributes change.

[0042] (3) Relationship mining: Combine rule base and machine learning algorithm to mine the correlation between entities, such as "user - use electricity - transformer", "transformer - by... power supply - transformer", "weather warning - influence - transformer", "policy - applicable to - user" and so on. Create relationship when new entity interaction event is identified, and update relationship attribute when relationship state changes (such as subscription taking effect / invalidation).

[0043] (4) Attribute completion: Use algorithm (such as TransE algorithm) to complete missing attributes based on the correlation between entities, such as completing the "high load" attribute label according to the transformer load rate historical data and current weather conditions.

[0044] (5) Knowledge graph storage: Use graph database (such as Neo4j / JanusGraph) to store entities, relationships and attributes, support efficient correlation query and reasoning. Write entity, relationship structure data into graph database through Flink Sink operator, and write weight data into time series database at the same time; Introduce Flink CEP operator to monitor and update abnormals, trigger automatic alarm and rollback, and ensure graph consistency.

[0045] Key weight parameters, adaptive adjustment algorithm is applied. The following example is a lightweight adaptive algorithm for user demand intensity weight (W demand ) and device risk level weight (W risk ), which takes into account both calculation efficiency and adjustment accuracy: (1) User demand intensity weight (W demand ) adjustment algorithm Core logic: Based on the weighted model of demand frequency + urgency + time decay, the value range is [0, 1], the formula is as follows: W demand =0.4*F freq +0.4*F urgency +0.2*

[0046] Parameter description: F freq : Normalized value of user demand frequency for target service in the last 24 hours (demand times / historical daily average times).

[0047] F urgency : Demand type weight coefficient (fault repair = 1.0, emergency consultation = 0.6, ordinary business = 0.2).

[0048] : Time decay factor, t is the interval (hours) between the current time and the last demand, the longer the interval, the more obvious the weight decay.

[0049] Execution mode: Flink real-time capture user demand events, trigger Ffreq and t update; daily calibration of weight distribution coefficient, ensure adaptation to user behavior changes.

[0050] 2) Equipment risk level weight (W risk ) adjustment algorithm Core logic: based on multi-factor weighted summation model, value range [0, 1], formula as follows: W risk =0.3*F dev +0.3*F history +0.2*F life +0.2*F node Parameter description: F dev : equipment real-time running parameter deviation normalization value (such as voltage / current deviation ratio of rated value): F history : normalized value of equipment failure frequency in the past 3 months.

[0051] F life : equipment service life ratio (service length / design life).

[0052] F node : importance coefficient of power grid node where the equipment is located (hub node = 1.0, branch node = 0.5).

[0053] Execution mode: Flink collects equipment running data every 5 minutes, updates F dev and calculates weight; based on fault data calibration factor weight every month, update risk threshold value synchronously, when W risk ≥0.7, trigger early warning push.

[0054] Step S3 generates multi-dimensional user portrait based on power domain knowledge graph and user power consumption behavior analysis. User portrait includes but is not limited to: Power consumption characteristics: including power consumption load level, power consumption peak period, power consumption fluctuation amplitude, new energy access situation, etc.

[0055] Service demand preference: including attention degree to electricity fee reminder, fault repair, power consumption optimization suggestion, policy subsidy and other services.

[0056] User type label: including residential users, small and micro enterprise users, industrial park users, high energy consumption enterprise users, etc.

[0057] Credit rating: credit score based on payment records, default situations and other data.

[0058] Step S4 formulates a personalized push strategy based on the power domain knowledge graph inference result, user portrait, and power grid operation state. Please combine Figure 2 The workflow is as follows: (1) Demand identification Integrate the knowledge graph inference result and user portrait, and perform knowledge graph association reasoning. For example, “transformer high load + user belongs to the transformer power supply area of the transformer → user has high load warning and power optimization demand”, “policy entity - applicable to - user → user has policy subsidy application demand”.

[0059] This step takes the power domain knowledge graph as the core carrier and realizes the mapping from multi-source data to user accurate demand through the closed-loop process of “equipment state quantification - entity association matching - demand confidence calculation - inference chain generation”. Taking “transformer high load + user belongs to the transformer power supply area of the transformer → user has high load warning and power optimization demand” as a typical scenario, the quantification calculation method and inference chain generation process are described in detail.

[0060] ① Transformer high load quantification index Define the real-time load rate L of the transformer T as the core quantification index of the equipment state, and the calculation formula is as follows: L T =P T实时 / P T额定 *100% P T实时 : Real-time active power of the transformer, obtained by the power Internet of Things collection terminal, with a collection frequency of 15 minutes / time and a data accuracy of ±2%.

[0061] P T额定 : Transformer rated capacity, stored in the knowledge graph “power equipment” entity attribute library, associated with the unique asset number of the equipment.

[0062] Set the high load judgment threshold interval (based on the power industry standard DL / T 572-2010):

[0063] When L T ≥80%, it is determined that the transformer enters a high load state, triggering the downstream associated reasoning process.

[0064] ② User-transformer area association quantification index Define the area affiliation correlation degree R U-T , which represents the membership relationship between the user and the power supply transformer, with the following value rules: R U-T =1: Indicates that the “user electricity meter → power distribution branch line → transformer outlet end” topology link is completely matched.

[0065] R U-T =0.5: indicates that the user's electric meter belongs to the fuzzy matching of the substation area (such as temporary power switching).

[0066] R U-T =0: indicates no clear topological association.

[0067] Source of association data: entity relationship chain of "user-smart meter-power distribution line-transformer" in knowledge graph, combined with power GIS topological data and marketing archive data cross verification, to ensure the accuracy of association ≥99.5%.

[0068] ③Demand confidence quantitative index Define demand confidence C D as a quantitative evaluation index of demand effectiveness, combining load rate, substation correlation degree, and high load duration, the calculation formula is as follows: C D =w1*L T,归一化 + w2*R U-T + w3*D T,归一化 Where: L T,归一化 : load rate normalized value, L T,归一化 =(L T -80%) / 20%, mapping 80%-100% interval to 0-1.

[0069] D T,归一化 : high load duration normalized value, D T,归一化 =t 持续 / t 阈值 , t 阈值 is set to 2 hours, t 持续 ≥2 hours take 1.

[0070] Weight coefficient: based on expert experience and historical data training in the field of electricity, w1=0.5, w2=0.3, w3=0.2, and w1+w2+w3=1.

[0071] Set confidence threshold: when C D ≥0.7, it is determined that the user has high load warning and electricity optimization demand.

[0072] ④Demand reasoning chain generation process The generation of reasoning chain is based on entity relationship traversal of knowledge graph and rule engine reasoning, which is divided into four core steps to form a traceable and explainable demand reasoning path, as follows: 1) Initialization of knowledge graph entities and relationships Extract three types of core entities and relationships from the knowledge graph in the field of electricity: Device entity: Transformer T (with attributes: asset number, rated capacity, real-time load rate, load state).

[0073] User entity: Power user U (with attributes: user ID, meter number, power type, substation ownership).

[0074] Relationship type: Power supply relationship (transformer → distribution line → user meter), ownership relationship (user → substation).

[0075] 2) High-load transformer entity screening A. Read transformer real-time load rate data and calculate L T .

[0076] B. According to the load state threshold, screen out the high-load transformer set T high ={T1,T2,...,T n}.

[0077] C. Label the transformers in the set with load state S2 or S3, and record the high-load duration t 持续 .

[0078] 3) User-transformer topology association matching A. Traverse all user entities U in the knowledge graph, and trace back their corresponding power supply transformers T through the power supply relationship U .

[0079] B. Determine whether T U belongs to the high-load transformer set T high .

[0080] C. Calculate the substation ownership association degree R U of the user and T U-T , and screen out the user set U U-T with R 关联 ≥0.5.

[0081] 4) Rule engine triggering and reasoning chain output A. Call the preset rule engine, input parameters: L T,归一化 , R U-T , D T,归一化 , calculate the demand confidence C D .

[0082] B. When C D ≥0.7, trigger the demand judgment rule and generate a standardized reasoning chain.

[0083] C. Output the explainable reasoning path in the following format: > Reasoning chain example: > Transformer T001 real-time load rate L T=89 % → Determined as high load state S2 → Electricity meter number M456 of user U123 is associated to T001 via distribution line L789 → Substation attribution association degree R U-T =1 → High load duration T 持续 =2.5 hours → Demand confidence C D =0.5*0.45+0.3*1+0.2*1=0.725 → Meets threshold 0.7 → Determines that user U123 has high load warning and electricity optimization demand.

[0084] 5) Reasoning chain storage and traceability The generated reasoning chain is bound to the user demand and stored in the knowledge graph "demand reasoning" entity attribute, supporting subsequent demand pushing strategy optimization, reasoning result audit and traceability.

[0085] (2) Scene matching Combine demand identification, power grid operation state and external associated data (such as weather condition information) to match the corresponding service scene. For example, "transformer load rate 85% (high load) + high temperature warning → high load warning scene". Take the typical scene as an example: ① Scene matching model specific implementation This module adopts a three-layer weighted scene matching model, based on the user demand confidence output by the demand identification module, power grid operation state data, and weather condition information, to construct a multi-dimensional feature vector, match through threshold determination and scene rule library, and output the target service scene. The core process is as follows: A. Feature normalization processing: Map power grid operation state indicators (such as transformer load rate 0-100%), weather condition parameters (such as temperature -20℃-50℃), and user demand confidence (0-1) to the 0-1 interval to generate a normalized feature set X=(x grid ,x weather ,x demand ).

[0086] B. Scene rule library construction: Pre-set power service scene rules, each rule contains feature combination conditions and matching scenes, for example:

[0087] Weighted matching degree calculation: Introduce feature weight coefficients W=(w grid ,w weather ,w demand ) (based on power business priority setting, meet the total of 1), calculate the scene matching degree M, the formula is as follows: M=w1*x1*r1+w2*x2*r2+w3*x3*r3 Where: r1, r2、 , r 3、 is the relevance of the feature to the scene (value 0-1, labeled by domain experts).

[0088] Scene output determination: Set the scene trigger threshold M threshold = 0.75, when M ≥ M threshold , match the corresponding service scene; when multiple scenes are met, select the scene with the highest matching degree as the output.

[0089] ② Influence mechanism of power grid operation data on push strategy As the core basis for decision-making, power grid operation data influences the push strategy from three dimensions: push priority, push content, and push timeliness, as follows: A. Push priority: Based on the abnormality level of power grid operation indicators, high abnormality level corresponds to high push priority. For example: transformer load rate above 90% (severe overload) → push priority is level one (immediate push); load rate 80%-90% (mild overload) → push priority is level two (push within 30 minutes).

[0090] B. Push content: Customize content according to the abnormal type of operation data. For example: high transformer load → push electricity optimization suggestions (peak-shaving time period, load sharing scheme); low line voltage → push voltage anomaly reminder (fault repair entry, temporary electricity use precautions).

[0091] C. Push timeliness: Dynamically adjust the push timing in combination with the real-time changes of operation data. For example: continuously rising load rate (rise by ≥2% every 5 minutes) → immediately trigger push; load rate stable at 85% and no upward trend → delay to non-peak electricity use period for push.

[0092] ③ Association between power grid operation state and weather conditions A. Association logic design Build power grid-weather association rules, the core logic is: weather conditions indirectly change power grid operation state by affecting user electricity use behavior, the coupling characteristics of the two trigger corresponding service scenes. Taking the "high load warning scene" as an example, the association logic chain is: high temperature weather → user high-power device usage rate increases (such as air conditioner) → transformer load rate rises → triggers high load warning scene.

[0093] B. Trigger threshold and quantification logic Set the coupling trigger threshold of power grid operation state and weather conditions. Taking the association between high temperature and high transformer load as an example, the specific quantification rules are as follows: a. Indicator threshold labeling Power grid operation indicator: transformer load rate threshold L high = 85% (load rate ≥ 85% is determined as high load).

[0094] Weather condition index: high temperature threshold T high = 38℃ (temperature ≥ 38℃ is determined as high temperature warning).

[0095] b. Coupling trigger logic When both "transformer load rate ≥ L high " and "temperature ≥ T high " are met, trigger the high load warning scenario.

[0096] Introduce the correlation influence coefficient k to correct the matching degree. When the weather condition aggravates the grid anomaly, k = 1.2 (improve the matching degree); when the weather condition has no influence, k = 1.0. The corrected matching degree formula is: M = M * k.

[0097] C. Reverse verification rule If the transformer load rate ≥ L high but the temperature < T high , then reduce the correlation influence coefficient (k = 0.8), and the matching degree M ≥ 0.8 can trigger the scenario to avoid misjudgment.

[0098] 3) Push parameter determination: according to the push channel preference (such as APP, short message, public number message) in the user portrait and the power grid operation state, determine the push content priority, push time window and push channel.

[0099] Step S5 pushes the power service information to the user terminal according to the push strategy, supporting multi-channel push: 1) Online channel: power APP, WeChat public number, short message, email, etc.

[0100] 2) Offline channel: smart meter display screen, community bulletin board (for residential users), enterprise dedicated customer manager docking (for enterprise users).

[0101] Push content example: High load warning scenario: "

Electric power service reminder

[0102] Policy subsidy declaration scenario: "

Policy subsidy notice

[0103] Further, the embodiment of the present application also collects user interaction feedback data, dynamically optimizes the knowledge graph and the push strategy: 1) Feedback data collection: Collect data such as clicks, views, collections, ignores, and feedback messages through user terminals, for example, a user clicking on the "electricity optimization suggestion" link is considered valid feedback, and multiple ignores of a certain type of push are considered no demand.

[0104] 2) Graph update: Regularly integrate new data and feedback data into the knowledge graph, update entity attributes (such as user demand preference labels) and relationships (such as the addition of "user-participation-demand response" relationships).

[0105] 3) Strategy optimization: Use reinforcement learning algorithms, with user interaction conversion rate (clicks / pushes) and user satisfaction score as reward functions, to optimize push content filtering rules, push timing judgment models, and channel selection strategies, for example, for users who have ignored multiple SMS pushes, adjust to APP push.

[0106] The hardware configuration example implemented by the embodiment of the application is as follows: Data collection uses edge computing gateway, supports multi-protocol access (Modbus, MQTT, HTTP), realizes real-time collection and transmission of multi-source data; Knowledge graph construction uses server cluster (CPU: Intel Xeon Gold 6248, memory: 128GB, hard disk: 2TB SSD), deploys Neo4j graph database and BERT+CRF model training environment; User portrait generation is based on Spark distributed computing framework, realizing parallel generation of large-scale user portraits; Push strategy decision uses GPU server (GPU: NVIDIA Tesla V100), deploys reinforcement learning model and associated reasoning engine; Information push execution is through the push server, supporting multi-channel push interface development and integration; Feedback optimization uses the stream processing framework Flink to realize real-time collection and processing of feedback data.

[0107] The software implementation method example is as follows: Operating system: Linux CentOS 7.9; Development language: Python 3.8, Java 11; Graph database: Neo4j 4.4; Machine learning framework: TensorFlow 2.8, PyTorch 1.12; Distributed computing framework: Spark 3.2, Flink 1.14; Interface development: Spring Boot 2.7, Django 4.0.

[0108] The application process of the power service information intelligent pushing method of the embodiment of the application is further illustrated by three cases.

[0109] Case One: Taking the "peak-valley electricity price adaptation + peak-shaving for saving money" pushing of a double-income family as an example.

[0110] (1) Case scenario Mr. and Mrs. Li, a double-income family in Futian District, Shenzhen, commute from 9 am to 5 pm, and there is only someone at home for electricity use after 19:00 at night and on weekends. They have been using the "Nanwang Online" APP to pay electricity bills for a long time. The system monitors that: their family's summer monthly electricity consumption is 380 degrees, the electricity consumption in the peak period (10:00-12:00, 14:00-19:00) accounts for only 12%, and the electricity consumption in the valley period (0:00-8:00) accounts for 45%. Currently, they are still implementing ordinary tiered electricity prices and have not opened the peak-valley electricity price service.

[0111] (2) Implementation process 1) Data collection: Through the "Nanwang Online" APP backend and electricity collection system, collect user basic data (double-income family, 2 people living), electricity behavior data (summer monthly electricity consumption of 380 degrees, valley period electricity consumption concentration), and electricity price policy data (Shenzhen peak-valley electricity price standard: peak period 1.1121 yuan / degree, flat period 0.6542 yuan / degree, valley period 0.2486 yuan / degree).

[0112] 2) Knowledge graph construction: extract entities (Mr. Li's family, a certain subdistrict in Futian District, peak-valley electricity price policy), mine relationships (Mr. Li's family - low peak period electricity consumption - adapt to peak-valley electricity price, high valley period electricity consumption - can save money through policy), and complete the "large peak-shaving electricity consumption potential" attribute label.

[0113] 3) User portrait generation: electricity characteristics (valley period electricity consumption dominant, peak period electricity consumption less), service demand preference (electricity bill saving, convenient handling), and user type label (double-income commuting family).

[0114] 4) Pushing strategy decision: identify core needs (peak-valley electricity price policy notification, cost savings calculation, online opening); match the "high valley period electricity consumption + peak-valley electricity price adaptation" scenario; determine the pushing parameters as: APP pop-up window + message center pushing, pushing time Friday 20:00 (weekend decision window period), and content highlighting cost savings comparison.

[0115] 5) Information push execution: APP pushes the content "

Peak and valley electricity price money saving reminder

[0116] 6) Feedback optimization: Mr. Li opens the peak and valley electricity price through the APP one key, and the system collects the successful feedback and subsequent electricity saving data, updates the knowledge graph "Mr. Li's family- has opened- peak and valley electricity price" relationship, and subsequently pushes "valley segment electricity consumption period suggestion" (such as charging at night, low valley washing).

[0117] (3) Case effect Mr. Li's family saves 72 yuan in the first month of peak and valley electricity price settlement, which effectively reduces the cost of life; the system accurately identifies suitable users through data, so that the peak and valley electricity price policy really benefits the target group, and guides users to staggered electricity consumption, helping to optimize the load of the power grid.

[0118] Case 2: Take the industrial park enterprise user (a certain technology company) as an example.

[0119] (1) Case scenario A certain technology company is located in the No. 1 block of the incubator, which is powered by transformer T102. In the summer high temperature weather, the transformer load rate continues to rise to 85% (high load), and the government issues a summer demand response subsidy policy. The company meets the declaration conditions.

[0120] (2) Implementation process 1) Data collection: The data collection module collects multi-source data, including the electricity consumption behavior data of a certain technology company (the recent electricity consumption peak period is 10:00-16:00), the user type label (industrial park enterprise), the load rate data of transformer T102 (85%), the high temperature early warning information, and the demand response subsidy policy data.

[0121] 2) Knowledge graph construction: The knowledge graph construction module extracts entities (a certain technology company, No. 1 block of the incubator, transformer T102, high temperature early warning, demand response subsidy policy), mines relationships (a certain technology company- electricity consumption in- No. 1 block of the incubator, No. 1 block of the incubator- powered by- transformer T102, high temperature early warning- influence- transformer T102, demand response subsidy policy- applicable to- a certain technology company), and completes the "high load" attribute of transformer T102.

[0122] 3) User portrait generation: Generate a user portrait of a certain technology company, including electricity characteristics (peak period 10:00-16:00, high energy-consuming enterprises), service demand preferences (focus on policy subsidies, electricity cost optimization), and user type labels (industrial park enterprises).

[0123] 4) Push strategy decision: The demand identification unit identifies the core needs of the company (high load warning, electricity optimization suggestion, demand response subsidy application) through association reasoning; the scene matching unit matches the "high load warning + policy subsidy application" composite scene; the push parameter determination unit determines the push channel as the power APP + enterprise dedicated customer manager interface according to the user portrait, the push time is 17:00 (avoiding the peak period of electricity consumption), and the push content priority is subsidy policy application > high load warning > electricity optimization suggestion.

[0124] 5) Information push execution: At 17:00, the integrated service information is pushed to the certain technology company through the power APP, and the dedicated customer manager reminds by phone.

[0125] 6) Feedback optimization: The certain technology company clicks on the APP push link to view the subsidy application details and completes the application. The feedback optimization module collects effective clicks and application behaviors, updates the "certain technology company-participates-demand response" relationship in the knowledge graph, and optimizes the push strategy for this type of enterprise user (subsequent priority for pushing policy subsidy information).

[0126] (3) Case effect Through the precise push of the system, the certain technology company timely understands the transformer high load risk and demand response subsidy policy, successfully participates in demand response and obtains subsidies, which not only reduces electricity costs, but also helps the power grid to alleviate power supply pressure, achieving a win-win situation for users and the power grid.

[0127] Case three: Take the "one household multiple population ladder expansion + convenient application" push of a 5-person family in Shenzhen City, Guangdong Province as an example.

[0128] (1) Case scenario Ms. Zhang's family in Shenzhen Bao'an District, with 5 people living together (husband and wife + 2 children + 1 elderly), inquires electricity charges through the "Nanwang Online" APP, and the system monitors that their summer electricity consumption has exceeded 400 degrees for nearly 6 months, has entered the second tier of the ladder for 3 consecutive months, and has not applied for the "one household multiple population" ladder electricity expansion policy, which meets the conditions of the Shenzhen 5-person and above household monthly electricity consumption increase of 100 kilowatt-hours in each tier.

[0129] (2) Implementation process 1) Data collection: Through the real-name authentication information of the "Southern Power Grid Online" APP and the electricity collection system, basic user data (5 people living together, household registration address and electricity address are consistent), electricity consumption behavior data (average monthly consumption of 420 kWh in summer, second-tier electricity consumption of 160 kWh), and policy data (Shenzhen's tiered expansion policy for households with multiple people, online linkage processing channels).

[0130] 2) Knowledge graph construction: Extract entities (Ms. Zhang's family, a certain power station in Bao'an District, the policy of one household with multiple people, and the residential information linkage service), and explore relationships (Ms. Zhang's family - meets the policy of one household with multiple people, the second level of high electricity consumption - can be reduced through capacity expansion, online linkage - no additional proof required).

[0131] 3) User profile generation: electricity usage characteristics (high load due to large population, frequent overspending in tiered pricing), service demand preferences (policy benefits, less legwork required for processing), user type tags (families with five or more people).

[0132] 4) Push strategy decision-making: Demand identification of core needs (policy notification, expansion benefits, joint processing); Scenario matching of "multi-population + tiered over-limit" composite scenarios; Push parameters are determined as: APP push + SMS reminder, push time Wednesday 10:00 (family service hours), content includes benefit calculation and joint processing entry (no need to open a separate residence certificate).

[0133] 5) Information Push Execution: The APP push notification reads: "[Electricity Bill Discount Reminder for Multiple Households] Ms. Zhang, your family of 5 lives together and you qualify for Shenzhen's tiered electricity capacity expansion policy! Currently, you pay an extra 80 yuan per month for the second tier of electricity bills. After applying, the electricity capacity for each tier will increase by 100 kWh, saving you an average of 56 yuan per month in the summer. Click 'Multiple Households' to apply. The application will be completed in 3 minutes and will take effect the following month!"

[0134] 6) Feedback optimization: Ms. Zhang completed her application through the "Residential Information + Electricity" linkage service and received a notification of approval 3 working days later. The system updated the knowledge graph relationship of "Ms. Zhang's family - already enjoying - multiple people in one household". Subsequently, a renewal reminder will be pushed two months before the policy expires.

[0135] (3) Case Study Results Ms. Zhang's family completed the entire process online without needing to prepare any additional materials. After enjoying the capacity expansion policy in the first month, they saved 59 yuan on electricity bills. The system solves the problem of "difficulty in obtaining proof" through government-enterprise data linkage, allowing the beneficial policies to reach users directly and improving user satisfaction with electricity services.

[0136] Corresponding to the intelligent power service information push method described in Embodiment 1 of the present invention, Embodiment 2 of the present invention also provides an intelligent power service information push device, comprising: The data acquisition module is used to collect multi-source heterogeneous data related to power services; a construction module configured to construct a power service knowledge graph based on multi-source heterogeneous data, the knowledge graph comprising user entities, device entities, service entities, power grid node entities, and relationships and dynamic weight parameters therebetween; a generation module configured to generate a user portrait based on the power service knowledge graph, the user portrait comprising power consumption characteristics, service demand preferences, user type labels, and credit levels; a decision module configured to determine a push strategy by performing demand identification and scenario matching based on the user portrait and power grid operation states through the power service knowledge graph; a push module configured to push power service information to a user terminal according to the push strategy.

[0137] Corresponding to the power service information intelligent push method of the foregoing embodiment one, the embodiment three further provides a power service information intelligent push device, comprising: one or more processors; a memory; one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, and the one or more application programs are configured to execute the power service information intelligent push method of the foregoing embodiment one.

[0138] Corresponding to the power service information intelligent push method of the foregoing embodiment one, the embodiment four further provides a computer program product comprising computer instructions, which instruct a computer device to perform operations corresponding to the power service information intelligent push method of the foregoing embodiment one.

[0139] Preferably, the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor can also be any conventional processor. The processor is the control center of the device, and connects various parts of the device through various interfaces and lines.

[0140] The memory mainly includes a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function, and the like, and the data storage area can store related data and the like. In addition, the memory can be a high-speed random access memory, and can also be a non-volatile memory such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, and the like, or the memory can also be other volatile solid-state storage devices.

[0141] It should be noted that the above device can include but is not limited to a processor and a memory, and those skilled in the art can understand it.

[0142] As can be seen from the above description, compared with the prior art, the beneficial effects of the present application are that the present application realizes the fundamental change of power service information push from passive response to active prediction by constructing an intelligent push strategy decision mechanism based on a power service knowledge graph. The mechanism innovatively integrates user portraits and real-time operation states of the power grid to establish a dynamic decision-making closed loop of demand identification and scene matching: first, the current power demand of the user is accurately captured, the potential demand is inferred in combination with the entity relationship in the knowledge graph, and the demand urgency and feasibility are evaluated according to real-time parameters such as power grid load rate, equipment availability, and fault influence range; second, the matching degree of user demand and service entity is calculated, the demand urgency, service availability, user preference, and power grid carrying capacity are comprehensively evaluated by a multi-dimensional weighted scoring model to realize adaptive generation of the push strategy; at the same time, the decision-making process is interpretable, the system automatically labels the push basis, and the push logic is transparent. The mechanism effectively solves the core problems of poor accuracy, information overload, and lack of association in traditional push, significantly improves the pertinence and initiative of information push, changes the service from after-the-fact remediation to pre-emptive prevention, greatly reduces the demand for manual intervention, builds an intelligent and interpretable precise push core capability for power enterprises, and comprehensively supports the transformation of power services to active, personalized, and intelligent.

[0143] The above disclosure is only the preferred embodiment of the present application, and of course cannot limit the scope of the right of the present application, therefore, the equivalent changes made according to the claims of the present application still belong to the scope covered by the present application.

Claims

1. A method for intelligently pushing electricity service information, characterized in that, include: Step S1: Collect multi-source heterogeneous data related to power services; Step S2: Construct a power service knowledge graph based on multi-source heterogeneous data. The knowledge graph includes user entities, equipment entities, service entities, power grid node entities, and the relationships and dynamic weight parameters between them. Step S3: Generate a user profile based on the power service knowledge graph. The user profile includes electricity consumption characteristics, service demand preferences, user type tags, and credit rating. Step S4: Based on the user profile and power grid operation status, the power service knowledge graph is used to identify needs and match scenarios to determine the push strategy; Step S5: Push the power service information to the user terminal according to the push strategy.

2. The method according to claim 1, characterized in that, In step S4, the requirement identification specifically includes: Define the core operational quantitative indicators of the target equipment, calculate the indicator values ​​based on the equipment's rated parameters and real-time operational data, set multiple state threshold ranges, and initiate correlation inference when the indicator value reaches the preset trigger threshold; Define the correlation parameter between users and target devices, determine the correlation value based on the entity relationship chain and cross-validation data in the power service knowledge graph, and filter the set of users that meet the correlation threshold. Construct a weighted calculation model that includes the quantified and normalized values ​​of device status, entity correlation, and normalized values ​​of status duration, determine the weight coefficients of each dimension, calculate the confidence level of the demand, and determine that the user has a corresponding demand when the confidence level reaches a set threshold. Based on the entity relationship traversal and rule engine reasoning of the power service knowledge graph, a traceable demand reasoning path is generated, and the reasoning chain is bound to and stored with user needs.

3. The method according to claim 2, characterized in that, In the quantification of equipment status, the calculation formula for the core operational quantification indicators is as follows: Index value = (Real-time operating parameters of equipment / Rated parameters of equipment) × 100% The real-time operating parameters of the equipment are acquired through the power Internet of Things acquisition terminal, and the state threshold range includes at least three levels: normal state, early warning state, and alarm state.

4. The method according to claim 2, characterized in that, The rules for determining the entity correlation degree are as follows: when the correlation degree = 1, it indicates that the topological link between the user and the target device is completely matched; when the correlation degree = 0.5, it indicates that the affiliation relationship between the user and the target device is fuzzy matched; when the correlation degree = 0, it indicates that there is no explicit topological association; the cross-validation data includes power GIS topological data and marketing archive data.

5. The method according to claim 2, characterized in that, The generation process of the inference chain includes: Extract target device entities, user entities, and their relationships from the power service knowledge graph to complete entity and relationship initialization; Filter the set of devices in an abnormal state, and mark the device status and duration of the abnormality; Iterate through user entities, trace their associated target devices, and filter the set of users that meet the association threshold. Call the rules engine, input the normalized value of the quantitative indicator, the correlation degree, and the normalized value of the duration, and calculate the demand confidence level; When the confidence level reaches the judgment threshold, a standardized reasoning path is generated, which is then bound to the user's needs and stored in the knowledge graph's requirement reasoning entity attributes.

6. The method according to claim 1, characterized in that, In step S4, the scene matching specifically includes: The power grid operation status indicators, meteorological condition parameters, and user demand confidence scores are mapped to the [0,1] interval to generate a normalized feature set; Pre-defined rules for power service scenarios are provided, with each rule containing a combination of feature conditions and the corresponding service scenario. Introduce feature weight coefficients to calculate scene matching degree; When the scene matching degree is greater than the preset scene triggering threshold, the corresponding service scene is matched; when multiple scenes are satisfied, the scene with the highest matching degree is selected as the output.

7. The method according to claim 6, characterized in that, The scene matching process also includes constructing power grid-meteorological association rules, specifically: Set a coupling trigger threshold between power grid operation indicators and meteorological conditions. The power grid operation indicators include at least equipment load rate and line voltage, and the meteorological conditions include at least high temperature and rainstorm. A correlation impact coefficient is introduced to correct the scenario matching degree. When meteorological conditions aggravate power grid anomalies, the correlation impact coefficient is greater than 1; when meteorological conditions have no impact, the correlation impact coefficient is 1. If the power grid operation indicators meet the threshold but the meteorological conditions do not, the correlation influence coefficient is less than 1. When the corrected scenario matching degree reaches the preset reverse verification threshold, the corresponding service scenario is triggered.

8. The method according to claim 1, characterized in that, In step S4, the push strategy includes push content priority, push time window and push channel. The push content priority is determined based on the user's demand confidence and the degree of power grid anomaly. The push time window is dynamically adjusted based on the user's peak electricity consumption period and the power grid operation status. The push channel is determined according to the push channel preference in the user profile.

9. The method according to claim 1, characterized in that, In step S4, the power grid operation data influences the push strategy from three dimensions: push priority, push content, and push timeliness. Specifically, this includes: The abnormality level of power grid operation indicators is classified into levels, and the higher the abnormality level, the higher the push priority. Corresponding push time limit requirements are set for different abnormality levels. Customized content based on the types of anomalies reflected in power grid operation data; The timing of push notifications is dynamically adjusted based on real-time changes in power grid operation data. Push notifications are triggered immediately when power grid operation indicators show a continuous deterioration trend, and when indicators are stable in the warning state and show no deterioration trend, push notifications are delayed until off-peak electricity consumption periods.

10. The method according to claim 1, characterized in that, Step S2 specifically includes: Extract the feature attributes of user entities, device entities, service entities, and power grid node entities; Determine the relationships between entities; Dynamic weight parameters are assigned to each relationship. These dynamic weight parameters are calculated based on the time decay factor, the association strength coefficient, and the user feedback weight.

11. The method according to claim 1, characterized in that, It also includes feedback optimization steps, specifically including: User interaction feedback data on push information is collected through user terminals. The interaction feedback data includes clicks, views, favorites, ignores, and feedback messages. Clicking on a link related to the target service is considered valid feedback, and ignoring a certain type of push multiple times is considered as having no corresponding need. Regularly integrate the newly added data and the aforementioned interactive feedback data into the power service knowledge graph, and update the entity attributes and relationships between entities in the graph; We employ reinforcement learning algorithms, using user interaction conversion rate and user satisfaction rating as reward functions, to optimize push content filtering rules, push timing judgment models, and push channel selection strategies.

12. The method according to claim 1, characterized in that, The multi-source heterogeneous data related to power services include user basic data, electricity consumption behavior data, power grid operation data, power business data, and external related data. Among them, electricity consumption behavior data includes electricity consumption, electricity consumption time period, electricity consumption mode, and electricity consumption anomaly records; power grid operation data includes power grid topology, equipment operating status, and fault records; power business data includes service request records, work order processing records, and service evaluation records; and external related data includes policy and regulatory texts, meteorological data, and socio-economic data.

13. A smart push device for electricity service information, characterized in that, include: The data acquisition module is used to collect multi-source heterogeneous data related to power services; The construction module is used to build a power service knowledge graph based on multi-source heterogeneous data. The knowledge graph includes user entities, equipment entities, service entities, power grid node entities, and the relationships and dynamic weight parameters between them. The generation module is used to generate user profiles based on the power service knowledge graph. The user profiles include electricity consumption characteristics, service demand preferences, user type tags, and credit ratings. The decision-making module is used to determine the push strategy by identifying needs and matching scenarios based on the user profile and the power grid operation status through the power service knowledge graph; The push module is used to push power service information to the user terminal according to the push strategy.

14. A smart push device for electricity service information, characterized in that, include: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to perform a smart push method for power service information as described in any one of claims 1 to 12.

15. A computer program product, characterized in that, Includes computer instructions that instruct a computer device to perform an operation corresponding to the method as described in any one of claims 1 to 12.

Images