Heating system production-charge-customer service integrated management method based on ai fusion

Abstract

The application discloses a heat supply system production-charge-customer service integrated management and control method fusing AI, and relates to the field of heat supply system management and control.The method comprises the following steps: constructing a heat user behavior track space-time prediction model through multi-source data, and pre-judging overdue payment, policy consultation, repair demand and complaint tendency behavior; establishing a heat user portrait model containing multi-dimensional labels such as payment ability and room temperature sensitivity; triggering a heat customer service intelligent agent and a heat operation and maintenance intelligent agent for single or composite behavior scenarios, and realizing intelligent payment urging, automatic policy reply, heat anomaly investigation, operation and maintenance work order distribution and personalized care pushing; further training a manual customer service intervention timing model, combining a heat manual customer service heterogeneous graph to construct an optimized distribution model, and realizing the precise distribution of the optimized manual customer service. The heat supply production, charge and customer service are integrated for management and control, and the traditional passive response service is changed into active prediction and personalized intelligent service.

Description

Technical Field

[0001] This invention belongs to the field of heating system operation and management technology, specifically involving an AI-integrated management and control method for heating system production, billing, and customer service. Background Technology

[0002] Heating customer service is the core link between heating companies and heat users, responding promptly to user requests for repairs, inquiries, and complaints, and quickly resolving issues such as insufficient room temperature, metering abnormalities, and payment disputes. In recent years, intelligent customer service has developed rapidly, replacing a large amount of repetitive work with human customer service representatives and reducing labor costs. However, with the increasing number of heat users and the diversification of service demands, the traditional intelligent customer service model is no longer adequate to meet the requirements of efficient, accurate, and user-friendly service. The main problems are as follows: 1) The existing heating customer service system is not integrated with the heating production system and the billing system. It relies more on the heat users to actively report repairs, complaints, or overdue payments before it begins to be handled. This makes it easy for heat users' problems to escalate, resulting in a poor heat user experience, high problem handling costs, and a high risk of group complaints. 2) Traditional hotline customer service often uses indiscriminate collection reminders and uniform response scripts, which cannot distinguish the core characteristics of hotline users such as room temperature sensitivity, communication preferences and complaint tendencies, which can easily cause hotline users to feel resentful and result in low accuracy of response strategies; 3) There is no connection between the heating customer service and the heating operation and maintenance side. When users have multiple behavioral needs at the same time, it often requires repeated communication with the heating users and multiple departments to handle the matter. The investigation of abnormalities and the soothing of emotions are not synchronized, which prolongs the time for problem resolution, easily leads to conflicts with heating users, and causes complaints to escalate. 4) The timing of intervention by heating customer service representatives often relies on the user's initiative to request human assistance or on the user's experience. This can lead to problems such as wasting manpower by intervening too early or causing complaints by intervening too late. In addition, the matching of heating customer service representatives with the user's behavioral needs does not take into account factors such as the suitability and workload of the customer service representatives, making it difficult to improve the problem resolution rate and user satisfaction.

[0003] Based on the aforementioned technical issues, a new integrated management and control method for heating system production, billing, and customer service that incorporates AI needs to be designed. Summary of the Invention

[0004] The technical problem to be solved by this invention is to overcome the shortcomings of the existing technology and provide an integrated management and control method for heating system production, billing and customer service that integrates AI. Through the complete link of heat user behavior prediction, heat user profiling, intelligent behavior scenario processing and human intervention optimization, it realizes integrated management and control of heating production, billing and customer service, transforming the traditional passive response service into proactive predictive and personalized intelligent service, and realizing intelligent, automated and personalized heating service throughout the entire process.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: This invention provides an integrated management and control method for heating system production, billing, and customer service that incorporates AI, comprising: The system obtains heat user payment data, repair data, customer service data, and heat metering data from the heating billing subsystem, customer service subsystem, and production operation subsystem to form a heat user behavior sequence. After extracting time features, behavioral features, and spatial correlation features from the heat user behavior sequence, a spatiotemporal prediction model of heat user behavior trajectory is constructed to predict the behavior of heat users in different future time periods, including overdue payment behavior, policy consultation behavior, repair request behavior, and complaint tendency behavior. Construct a user profile model that includes multi-dimensional labels such as user payment ability, room temperature sensitivity, communication preferences, and complaint tendency; When a heating user only has overdue payment behavior in the future, the heating customer service intelligent agent will be automatically triggered. Based on the heating user profile model, the agent will analyze the heating user's willingness to pay and communication preferences. The digital human customer service will then call the preset overdue payment knowledge base, match the heating user's payment collection strategy, and carry out intelligent collection of overdue heating users. When a heating user only inquires about policies in the future, the heating customer service AI will be automatically triggered, and the digital human customer service will extract the hot keywords of the inquiry and automatically match the relevant policy content to reply. When a heat user only has a need for repair and / or a tendency to complain in the future, the heating operation and maintenance intelligent agent will be automatically triggered to conduct intelligent investigation of heating anomalies and dispatch work orders to operation and maintenance personnel. At the same time, the heating customer service intelligent agent will be triggered to analyze the heat user complaint tendency score and the heat user room temperature sensitivity score based on the heat user profile model, and the digital human customer service will push personalized care messages based on the complaint tendency score. When a heating user defaults on their payment and has a need for repairs or a tendency to complain, the heating operation and maintenance intelligent agent is automatically triggered to conduct intelligent investigation of heating anomalies and dispatch work orders to operation and maintenance personnel. At the same time, the heating customer service intelligent agent is triggered to analyze the heating user's payment willingness score, communication preference, complaint tendency score, and room temperature sensitivity score based on the heating user profile model. The digital human customer service calls up the preset overdue payment knowledge base, matches the heating user's fee deferral strategy, prompts the heating user to apply for deferral, and promotes personalized care messages. Based on the historical dialogue data between digital human customer service and heat users, the heat user profile model, and the human customer service intervention data, a human customer service intervention timing model is trained to obtain the optimal timing for human customer service intervention. Then, combined with the constructed heterogeneous graph structure of heating human customer service, an optimized allocation model for heating human customer service intervention is established to allocate the optimal heating human customer service for dialogue services.

[0006] Furthermore, the data obtained from the heating billing subsystem includes: basic information of the heating user, heating payment time, payment amount, arrears duration, payment method, arrears frequency, and tiered heating consumption; the data obtained from the customer service subsystem includes: repair report data, fault type, fault location, and processing time; the time, type, content, processing result, and heating user sentiment of customer service inquiries and complaints; and the data obtained from the production and operation subsystem includes: supply and return water temperature and pressure of the heat exchange station, the heat exchange station to which the heating user belongs, pipeline zone, heat leakage rate of the area to which the heating user belongs, and fault occurrence rate. The formation of the heat user behavior sequence includes: using the heat user's unique account number as an identifier and using time period as the granularity, forming a time-series and structured heat user behavior sequence from heat payment data, repair data, customer service data, and heat metering data.

[0007] Furthermore, the extraction of time features from the hot user behavior sequence includes: using a short-term local time-series feature extraction module to extract short-term local time feature vectors of hot user behavior; using a long-term time-series dependency learning module to process the short-term local time feature vectors, learn the long-term time-series dependencies of hot user behavior, and output long-term time feature vectors; and using a dynamic weight fusion module to adaptively allocate weights to the short-term local time features and long-term time feature vectors, and fuse them to obtain a time feature vector. The extraction of behavioral features from the hot user behavior sequence includes: using an attention-enhanced feedforward neural network to perform nonlinear mapping on the hot user behavior sequence, mining the influence of each data in the behavior sequence on the hot user behavior trajectory, as well as the coupling relationship between each behavior, and outputting a behavioral feature vector; The spatial correlation feature extraction of the heat user behavior sequence includes: using heat users, heat exchange stations, and pipeline zones as graph nodes, and using the subordinate and adjacent relationships between heat users and heat exchange stations, users in the same pipeline zone, and heat exchange stations and associated pipelines as edges to construct a heating spatial topology graph; generating an adjacency matrix based on the heating spatial topology graph and performing normalization processing to eliminate the feature value deviation between the central node and the edge node to obtain a normalized adjacency matrix; using the heating spatial topology graph and the normalized adjacency matrix as input, using a graph convolutional neural network (GCN) to aggregate spatial features of the static spatial attributes and regional correlation behaviors of heat users, learning local spatial correlation and global spatial topology, and outputting an initial spatial feature vector; and performing masking optimization on the initial spatial feature vector based on the subordinate relationships of the heating spatial topology graph to mask the relevant features of non-associated heat exchange stations and pipeline zones, retaining key spatial features, and obtaining the final spatial correlation feature vector. Furthermore, the construction of a spatiotemporal prediction model for heat user behavior trajectories, predicting the behavior of heat users at different future time periods, includes: fusing temporal features, behavioral features, and spatial correlation features on a single user and its local space as units, outputting local spatiotemporal behavioral features; performing probability distribution learning on the local spatiotemporal behavioral features to mine the spatiotemporal behavioral correlation within the global heating supply scope, outputting global spatiotemporal behavioral features; using the global spatiotemporal behavioral features as input, employing a network structure with a fully connected layer and a multi-task output layer, performing feature learning and model training, constructing a spatiotemporal prediction model for heat user behavior trajectories, outputting the probability values ​​of a single heat user's overdue payment behavior, policy consultation behavior, repair request behavior, and complaint tendency behavior at different future time periods, and simultaneously outputting regional-level heat user behavior prediction data; wherein, the repair request behavior refers to heat user repair requests due to equipment or pipeline failures or substandard room temperature; the complaint tendency behavior refers to heat user complaints due to substandard room temperature or unsatisfactory fault handling.

[0008] Furthermore, the static spatial attributes include: physical location attributes, pipeline network affiliation attributes, and building attributes: the physical location attributes include the building, unit, floor, and latitude and longitude coordinates of the heat user; the pipeline network affiliation attributes include the heat exchange station, heating pipeline, and heating station zone to which the heat user belongs; the building attributes include the building's age, insulation level, and apartment layout to which the heat user belongs. The regional association behaviors include: neighborhood association behaviors, pipeline topology association behaviors, and regional aggregation behaviors; the neighborhood association behaviors include repair requests, complaints, and overdue payment behaviors of other users in the same building, unit, and heat exchange station; the pipeline topology association behaviors include abnormal data on pressure, temperature, and flow of the pipeline where the user is located; the regional aggregation behaviors include the overdue payment rate, repair frequency, complaint density, and the supply and return water temperature, pressure, and energy consumption indicators of the corresponding heat exchange station in the area.

[0009] Furthermore, the hot user payment ability label is based on the hot user's historical number of arrears, the proportion of arrears period, and the payment timeliness rate, and uses a weighted scoring method to quantify the hot user's willingness to pay. The room temperature sensitivity label is based on the frequency of room temperature repair requests, the frequency of room temperature inquiries, and the timeliness of feedback on room temperature fluctuations. A sensitivity scoring model is constructed using a gradient boosting tree to obtain the room temperature sensitivity score of heat users. The communication preference tags represent the user's preferences for communication channels, communication duration, communication style, and communication response. The complaint tendency label is quantified by a weighted scoring method based on the number of historical complaints, satisfaction with repair handling, percentage of negative emotions in conversations, percentage of unresolved issues, and complaint density in the same building.

[0010] Furthermore, when a heating user only has overdue payment behavior in the future, the heating customer service intelligent agent is automatically triggered. Based on the heating user profile model, the agent analyzes the user's willingness to pay and communication preferences. The digital human customer service then calls upon a preset overdue payment knowledge base to match the user's payment collection strategy and intelligently collect overdue payments from the user, including: When the probability of a heating user's overdue payment behavior is greater than the preset threshold, and the probabilities of policy consultation behavior, repair request behavior, and complaint tendency behavior are all less than the preset threshold, it is determined that only overdue payment behavior has occurred, and the heating customer service intelligent agent is automatically triggered. The heating customer service intelligent agent dispatches the heat user profile model to obtain the heat user's payment willingness score and communication preference characteristics. It prioritizes matching the strategies in the preset overdue payment knowledge base according to the payment willingness level, core communication channels and payment ability tags. If a complete match cannot be made, it performs fuzzy adaptation according to the payment willingness level and outputs a collection instruction that includes communication channels, contact time, core strategies and script templates. The digital human customer service team intelligently reminds users with outstanding fees to pay according to the collection instructions. Furthermore, when a heat user only inquires about policies in a future time period, the heating customer service intelligent agent will be automatically triggered, and the digital human customer service will extract the hot keywords of the inquiry and automatically match the relevant policy content to respond, including: When the probability of a heating user inquiring about policies is greater than a preset threshold, and the probabilities of overdue payments, repair requests, and complaints are all less than the preset threshold, it is determined that only policy inquiries have occurred, and the heating customer service AI is automatically triggered. The heating customer service intelligent agent calls the digital human customer service to preprocess the policy consultation text of heating users. After extracting the hot words of policy consultation through word segmentation and keyword extraction algorithms, it calls the heating policy knowledge base, performs precise matching or semantic similarity fuzzy matching based on the hot words of consultation, obtains the corresponding policy content, and organizes it into a structured reply text to reply to heating users.

[0011] Furthermore, when a heat user only requests repairs and / or has a tendency to complain in the future, the heating operation and maintenance intelligent agent is automatically triggered to conduct intelligent investigation of heating anomalies and dispatch work orders to operation and maintenance personnel. Simultaneously, the heating customer service intelligent agent is triggered to analyze the heat user's complaint tendency score and room temperature sensitivity score based on the heat user profile model. The digital human customer service representative then pushes personalized care messages based on the complaint tendency score, including: The heating operation and maintenance intelligent agent is triggered when the probability of a heat user reporting a repair request is greater than a preset threshold and the probability of other behaviors is less than a preset threshold; the heating operation and maintenance intelligent agent is triggered when the probability of a heat user complaining is greater than a preset threshold and the probability of other behaviors is less than a preset threshold; the heating operation and maintenance intelligent agent is triggered when the probability of both heat user reporting a repair request and complaining is greater than a preset threshold and the probability of other behaviors is less than a preset threshold. The heating operation and maintenance intelligent agent calls the built-in fault analysis model to analyze the operation data of the pipeline network and heat exchange station to which the target heat user belongs, as well as the heat metering data of the heat user's building or household. After analyzing the list of suspected fault triggering factors, it calls the built-in temperature difference analysis model to compare and analyze the temperature of the target heat user's community, building, and floors above and below, locate the root cause of the lack of heat, and generate a heating anomaly investigation report that includes the scope of the fault's impact and the root cause analysis report. The heating operation and maintenance intelligent agent intelligently matches the optimal operation and maintenance personnel based on the anomaly investigation report and the load of operation and maintenance personnel, and generates operation and maintenance work orders for dispatch. After the heating operation and maintenance intelligent agent investigates heating anomalies and dispatches operation and maintenance work orders, the heating customer service intelligent agent calls the heat user profile model to obtain the heat user complaint tendency score, heat user room temperature sensitivity score, and heat user communication preference. Based on the complaint tendency score and heat user room temperature sensitivity score, it classifies the users and calls the customer service care script knowledge base to match the corresponding care strategy. The digital human customer service pushes personalized care scripts according to the heat user communication preferences, and simultaneously informs the operation and maintenance work order dispatch status, the root cause of the fault or lack of heating, and the processing time.

[0012] Furthermore, when a heating user defaults on their payment and exhibits repair requests or a tendency to complain, the heating operation and maintenance intelligent agent is automatically triggered to conduct intelligent troubleshooting of heating anomalies and dispatch work orders to maintenance personnel. Simultaneously, the heating customer service intelligent agent is triggered to analyze the user's payment willingness score, communication preferences, complaint tendency score, and room temperature sensitivity score based on the user profile model. The digital human customer service then calls upon a pre-set overdue payment knowledge base to match the user's payment deferral strategy, prompting the user to apply for deferral and promoting personalized care messages, including: When the probability of a heating user's overdue payment is greater than a preset threshold, and the probability of a repair request or complaint is greater than a preset threshold, and the probability of a policy consultation is less than a preset threshold, the heating operation and maintenance intelligent agent is triggered. The heating operation and maintenance intelligent agent calls the built-in fault analysis model to analyze the operation data of the pipeline network and heat exchange station to which the target heat user belongs, as well as the heat metering data of the heat user's building or household. After analyzing the list of suspected fault triggering factors, it calls the built-in temperature difference analysis model to compare and analyze the temperature of the target heat user's community, building, and floors above and below, locate the root cause of the lack of heat, and generate a heating anomaly investigation report that includes the scope of the fault's impact and the root cause analysis report. The heating operation and maintenance intelligent agent intelligently matches the optimal operation and maintenance personnel based on the anomaly investigation report and the load of operation and maintenance personnel, and generates operation and maintenance work orders for dispatch. After the heating operation and maintenance intelligent agent investigates heating anomalies and dispatches operation and maintenance work orders, the heating customer service intelligent agent calls the heat user profile model to obtain the heat user's willingness to pay score, complaint tendency score, room temperature sensitivity score, and communication preference. Based on the willingness to pay classification, it matches the strategies in the preset overdue payment knowledge base and outputs the heat user's deferred payment strategy. At the same time, based on the complaint tendency score and the room temperature sensitivity score, it classifies and calls the customer service care script knowledge base to match the corresponding care strategy. The digital human customer service pushes the deferred payment application prompt and personalized care script according to the heat user's communication preference, and simultaneously informs the operation and maintenance work order dispatch status, the root cause of the fault or lack of heating, and the processing time.

[0013] Furthermore, based on the historical dialogue data between the digital human customer service representative and popular users, the popular user profile model, and the human customer service intervention data, a human customer service intervention timing model is trained to obtain the optimal timing for human customer service intervention in the dialogue, including: Acquire historical dialogue data between digital human customer service and popular users, popular user profile models, and human customer service intervention data as training sets; A human customer service intervention timing model is constructed, including a text feature extraction module, a profile feature fusion module, and an intervention timing prediction module. The training set is input into the human customer service intervention timing model, and the intervention accuracy and timing fit rate are used as training objectives to train the human customer service intervention timing model. During the interaction between the digital human customer service representative and the hot user, the dialogue data is collected in real time and the corresponding hot user profile data is retrieved. This data is then input into the trained human customer service intervention timing model, which outputs the probability of human customer service intervention and the optimal intervention time. When the intervention probability is greater than a preset threshold, the human customer service representative is automatically triggered to intervene in the dialogue at the optimal intervention time. The dialogue history and user profile information are also pushed to the human customer service representative for response.

[0014] Furthermore, based on the constructed heterogeneous graph structure of the heating customer service system, an optimized allocation model for heating customer service intervention is established, including: Based on historical processing data, business capability tags, performance scores and response time of heating customer service, a profile of heating customer service personnel is constructed, including professional capability score, communication style, load status and preferred heating user behavior scenarios. A heterogeneous graph of heating customer service is constructed with heat users, heating customer service representatives, and heat user behavior scenarios as nodes. At the same time, service association edges, capability matching edges, and load association edges are defined by combining heating customer service profiles, heat user profiles, and heat user behavior scenarios. The heterogeneous graph of heating human customer service is input into the heterogeneous graph attention network. The features of the heterogeneous graph are aggregated through node-level attention and semantic-level attention, and the feature vectors of heat user behavior demand and heating human customer service capability are output. Using the feature vectors of heat user behavior and the feature vectors of heating human customer service capabilities as input data, and with the objectives of maximizing matching degree, minimizing service cost, and optimizing response time, and constrained by the load, response time, and business capabilities of heating human customer service, an optimal allocation model for the intervention of heating human customer service is established to obtain the best heating human customer service.

[0015] The beneficial effects of this invention are: (1) This invention achieves multi-time period probability prediction of four types of behaviors of heat users in the future: overdue payment, policy consultation, repair needs and complaint tendency by integrating multi-source data of payment, repair, customer service and heat metering and extracting spatiotemporal features. It can identify potential service risks and needs in advance, upgrade passive response to proactive prediction and proactive communication when necessary, provide decision-making basis for intelligent collection, policy consultation, repair and complaint pre-processing, avoid the escalation of problems, and quantify the behavior pattern of heat users, providing data support for heating resource scheduling, customer service manpower allocation and charging strategy adjustment; (2) This invention transforms hot user behavior into four quantifiable tags: payment ability, room temperature sensitivity, communication preference, and complaint tendency, providing core feature inputs for subsequent intelligent payment reminders, care script push, and human customer service matching, making the service more in line with the actual situation of users; and through tags such as complaint tendency and room temperature sensitivity, it can quickly identify high-risk users and intervene in advance to reduce the complaint rate. (3) This invention addresses intelligent collection of overdue payments by implementing personalized collection strategies based on payment willingness scores and communication preferences, thereby increasing collection rates while reducing user resentment and labor costs. For policy consultations, it utilizes hot topic extraction and policy knowledge base matching to provide automatic answers, reducing the burden on human customer service, accurately matching policy content, and improving the accuracy of consultation responses and user satisfaction. For repair and complaint scenarios, the heating operation and maintenance intelligent agent proactively checks for anomalies and automatically dispatches orders, shortening fault response and processing time. The heating customer service intelligent agent then pushes personalized information based on complaint tendency scores and room temperature sensitivity. Personalized and caring communication techniques are used to soothe the emotions of heating users, reduce the risk of escalating complaints, and achieve a closed-loop service of rapid handling of faults and early emotional reassurance through collaboration between operation and maintenance and customer service. For scenarios involving overdue payments, repair requests, and complaints, deferred payment policies are pushed out based on payment willingness and repair request / complaint scenarios. This not only ensures that heating companies receive payments but also alleviates the payment pressure on heating users. For heating users who are sensitive to high room temperature and have a high tendency to complain, repair requests and complaints are handled first, while deferred payment plans are provided to avoid escalating conflicts. Ultimately, while ensuring the basic heating rights of heating users, the risk of arrears and complaints is minimized. (4) The present invention trains a model for the timing of human customer service intervention to obtain the optimal timing for human customer service intervention in dialogue, avoiding the waste of resources due to premature intervention by heating human customer service and the escalation of complaints due to late intervention; based on the heterogeneous graph structure, combined with customer service profiles, user profiles and behavioral scenario features, it achieves accurate matching of heating human customer service, enabling heating human customer service to effectively exert its business capabilities, balance customer service load, quickly solve heat user problems, and improve service efficiency and customer service resource utilization.

[0016] Other features and advantages will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart of an AI-integrated management and control method for a heating system that combines production, billing, and customer service. Figure 2 This is a flowchart of the method for extracting time features from hot user behavior sequences according to the present invention; Figure 3 This is a flowchart of the method for extracting spatial correlation features from hot user behavior sequences according to the present invention; Figure 4 The flowchart of the method for constructing a spatiotemporal prediction model of hot user behavior trajectory in this invention is shown below; Figure 5 This is a flowchart of the method for optimizing the timing of human customer service intervention in this invention; Figure 6 This is a flowchart of the heating system human customer service intervention optimization allocation method of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figure 1 As shown, this embodiment provides an integrated management and control method for heating system production, billing, and customer service that incorporates AI, including: The system obtains heat user payment data, repair data, customer service data, and heat metering data from the heating billing subsystem, customer service subsystem, and production operation subsystem to form a heat user behavior sequence. After extracting time features, behavioral features, and spatial correlation features from the heat user behavior sequence, a spatiotemporal prediction model of heat user behavior trajectory is constructed to predict the behavior of heat users in different future time periods, including overdue payment behavior, policy consultation behavior, repair request behavior, and complaint tendency behavior. Construct a user profile model that includes multi-dimensional labels such as user payment ability, room temperature sensitivity, communication preferences, and complaint tendency; When a heating user only has overdue payment behavior in the future, the heating customer service intelligent agent will be automatically triggered. Based on the heating user profile model, the agent will analyze the heating user's willingness to pay and communication preferences. The digital human customer service will then call the preset overdue payment knowledge base, match the heating user's payment collection strategy, and carry out intelligent collection of overdue heating users. When a heating user only inquires about policies in the future, the heating customer service AI will be automatically triggered, and the digital human customer service will extract the hot keywords of the inquiry and automatically match the relevant policy content to reply. When a heat user only has a need for repair and / or a tendency to complain in the future, the heating operation and maintenance intelligent agent will be automatically triggered to conduct intelligent investigation of heating anomalies and dispatch work orders to operation and maintenance personnel. At the same time, the heating customer service intelligent agent will be triggered to analyze the heat user complaint tendency score and the heat user room temperature sensitivity score based on the heat user profile model, and the digital human customer service will push personalized care messages based on the complaint tendency score. When a heating user defaults on their payment and has a need for repairs or a tendency to complain, the heating operation and maintenance intelligent agent is automatically triggered to conduct intelligent investigation of heating anomalies and dispatch work orders to operation and maintenance personnel. At the same time, the heating customer service intelligent agent is triggered to analyze the heating user's payment willingness score, communication preference, complaint tendency score, and room temperature sensitivity score based on the heating user profile model. The digital human customer service calls up the preset overdue payment knowledge base, matches the heating user's fee deferral strategy, prompts the heating user to apply for deferral, and promotes personalized care messages. Based on the historical dialogue data between digital human customer service and heat users, the heat user profile model, and the human customer service intervention data, a human customer service intervention timing model is trained to obtain the optimal timing for human customer service intervention. Then, combined with the constructed heterogeneous graph structure of heating human customer service, an optimized allocation model for heating human customer service intervention is established to allocate the optimal heating human customer service for dialogue services.

[0022] It should be noted that this application targets intelligent payment reminders, anomaly detection, and personalized care strategies designed for common heat user behavior scenarios. In practical applications, other behavioral scenarios may also be included, allowing for adaptive strategy design. The heating customer service intelligent agent and the heating operation and maintenance intelligent agent are core collaborative intelligent units, focusing respectively on heat user interaction and service strategies, and heating anomalies and resource scheduling. Through event triggering and data exchange, they form a complete intelligent service closed loop. The heating customer service intelligent agent, as the intelligent interaction hub on the user side, is responsible for responding to various user service needs, executing personalized service strategies, and managing the collaborative scheduling of digital human customer service and human customer service. It is the core bridge connecting users, the heat user profile model, the business knowledge base, and the human customer service system. Internally, it includes a trigger perception module, a profile invocation module, a strategy matching module, an interaction scheduling module, and an interaction control module. The trigger perception module receives real-time output from the behavior prediction model, identifies user behavior scenarios such as overdue payments, policy inquiries, repair requests, and complaints, and triggers corresponding service processes according to preset thresholds. The profile invocation module invokes the heat user profile model to extract user payment willingness scores, complaint tendency scores, room temperature sensitivity scores, and communication... Core features such as user preferences provide the basis for strategy matching; the strategy matching module is used to connect with business knowledge bases (overdue payment knowledge base, policy knowledge base, care script knowledge base, etc.), and match personalized service strategies (such as collection reminder strategy, deferred payment strategy, care script) based on user profiles and scenario characteristics; the interaction scheduling module is used to schedule digital human customer service to execute automatic replies, collection reminders and care pushes, and when the timing of human customer service intervention is triggered, it calls the human customer service intervention optimization allocation model, matches the best human customer service and pushes the work order; the interaction control module is used to monitor the entire interaction process between digital human customer service, human customer service and hot users, handle hot user inquiries and emotional feedback, record interaction data, and synchronously update user profiles and related knowledge bases.

[0023] The heating operation and maintenance intelligent agent, as the intelligent dispatch hub on the operation and maintenance side, is responsible for intelligent investigation of heating anomalies, dynamic allocation of operation and maintenance resources, and full lifecycle management of work orders. It is the core dispatch unit connecting the heating production and operation system, heat user repair reports and complaints, and operation and maintenance personnel. Internally, it is equipped with an anomaly perception module, an intelligent investigation module, a work order generation module, and a resource dispatch module. The anomaly perception module is used to receive repair and complaint trigger signals from the behavior prediction model, or to collect anomaly alarm data from the heating production and operation system, such as pipeline network, heat exchange station, and heat metering. The intelligent investigation module is used to perform anomaly identification, temperature difference analysis, etc., and, combined with heat user room temperature sensitivity scores, to locate the root cause of the anomaly and generate an investigation report. The work order generation module is used to generate operation and maintenance work orders based on faults, heat user locations, and root causes of anomalies. The resource dispatch module is used to match the optimal operation and maintenance personnel, dispatch work orders, and mark users with high room temperature sensitivity as high-priority work orders.

[0024] In this embodiment, the acquisition of heat user payment data from the heating billing subsystem includes: basic information of the heat user, payment time, payment amount, arrears duration, payment method, arrears frequency, and tiered heat consumption; the acquisition of heat user repair data and customer service data from the customer service subsystem includes: repair time, fault type, fault location, and processing time; time, type, content, processing result, and heat user sentiment of customer service inquiries and complaints; and the acquisition of heat user heat metering data from the production and operation subsystem includes: supply and return water temperature and pressure of heat exchange stations, heat exchange station to which the heat user belongs, pipeline zone, heat leakage rate of the area to which the heat user belongs, and fault occurrence rate. The formation of the heat user behavior sequence includes: using the heat user's unique account number as an identifier and using time period as the granularity, forming a time-series and structured heat user behavior sequence from heat payment data, repair data, customer service data, and heat metering data.

[0025] like Figure 2 , Figure 3 , Figure 4 As shown, in this embodiment, the extraction of time features from the hot user behavior sequence includes: using a short-term local time-series feature extraction module to extract short-term local time feature vectors of hot user behavior; using a long-term time-series dependency learning module to process the short-term local time feature vectors, learn the long-term time-series dependencies of hot user behavior, and output long-term time feature vectors; and using a dynamic weight fusion module to adaptively allocate weights to the short-term local time features and long-term time feature vectors, and fuse them to obtain a time feature vector. The extraction of behavioral features from the hot user behavior sequence includes: using an attention-enhanced feedforward neural network to perform nonlinear mapping on the hot user behavior sequence, mining the influence of each data in the behavior sequence on the hot user behavior trajectory, as well as the coupling relationship between each behavior, and outputting a behavioral feature vector; The spatial correlation feature extraction of the heat user behavior sequence includes: using heat users, heat exchange stations, and pipeline zones as graph nodes, and using the subordinate and adjacent relationships between heat users and heat exchange stations, users in the same pipeline zone, and heat exchange stations and associated pipelines as edges to construct a heating spatial topology graph; generating an adjacency matrix based on the heating spatial topology graph and performing normalization processing to eliminate the feature value deviation between the central node and the edge node to obtain a normalized adjacency matrix; using the heating spatial topology graph and the normalized adjacency matrix as input, using a graph convolutional neural network (GCN) to aggregate spatial features of the static spatial attributes and regional correlation behaviors of heat users, learning local spatial correlation and global spatial topology, and outputting an initial spatial feature vector; and performing masking optimization on the initial spatial feature vector based on the subordinate relationships of the heating spatial topology graph to mask the relevant features of non-associated heat exchange stations and pipeline zones, retaining key spatial features, and obtaining the final spatial correlation feature vector. Furthermore, the construction of a spatiotemporal prediction model for heat user behavior trajectories, predicting the behavior of heat users at different future time periods, includes: fusing temporal features, behavioral features, and spatial correlation features on a single user and its local space as units, outputting local spatiotemporal behavioral features; performing probability distribution learning on the local spatiotemporal behavioral features to mine the spatiotemporal behavioral correlation within the global heating supply scope, outputting global spatiotemporal behavioral features; using the global spatiotemporal behavioral features as input, employing a network structure with a fully connected layer and a multi-task output layer, performing feature learning and model training, constructing a spatiotemporal prediction model for heat user behavior trajectories, outputting the probability values ​​of a single heat user's overdue payment behavior, policy consultation behavior, repair request behavior, and complaint tendency behavior at different future time periods, and simultaneously outputting regional-level heat user behavior prediction data; wherein, the repair request behavior refers to heat user repair requests due to equipment or pipeline failures or substandard room temperature; the complaint tendency behavior refers to heat user complaints due to substandard room temperature or unsatisfactory fault handling.

[0026] It should be noted that the short-term local temporal feature module can employ a convolutional neural network, an InceptionTime network, or a local window Transformer encoder; the long-term temporal dependency learning module can employ a gated recurrent unit (GRU) or a full-sequence Transformer encoder. The dynamic weight fusion module uses an attention mechanism to dynamically assign weights to the short-term local temporal feature vector and the long-term temporal feature vector.

[0027] The heating spatial topology graph is the fundamental data structure for spatial feature aggregation in graph convolutional neural networks (GNNs). Without it, it's impossible to define which nodes need feature aggregation and how to allocate aggregation weights. Nodes carry the attribute features of heat users, heat exchange stations, and pipeline zones, while edges define the feature propagation path and determine the neighborhood range during GNN aggregation. The normalized adjacency matrix is ​​the core parameter for convolution operations in GNNs, controlling the propagation intensity of spatial features. GNNs have two convolutional layers. The first layer aggregates features of directly adjacent nodes, such as users in the same unit or belonging to the same heat exchange station, learning local spatial relationships. The second layer aggregates the neighbor features of neighbors, such as other pipeline heat users associated with the heat exchange station, belonging to the same community or the same road segment, learning the global spatial topology.

[0028] During mask optimization, weakly correlated features are determined based on the hierarchical relationships in the heating spatial topology. Node features that do not belong to the current user's heat exchange station or pipeline zone are identified as weakly correlated and masked. The topology is the basis for distinguishing between correlated and irrelevant spatial features. Attention-enhanced feedforward neural networks introduce an attention mechanism, assigning dynamic weights to different behavioral features for more accurate mining of coupling relationships. Global spatiotemporal behavioral correlation mining uses local spatiotemporal behavioral features as input to mine cross-regional and cross-user spatiotemporal behavioral correlations across the entire heating domain. Probabilistic distribution learning is used to model the probabilistic distribution of local spatiotemporal behavioral features (such as Gaussian Mixture Models (GMM) or Variational Autoencoders (VAEs)) to learn the distribution patterns of behaviors in different regions and heat users. A network structure with a fully connected layer and a multi-task output layer is adopted: feature depth mining is performed through the fully connected layer, and four independent output layers are set up to predict overdue payment behavior, policy consultation behavior, repair request behavior, and complaint tendency behavior respectively. Each layer uses a sigmoid activation function to output a probability value between 0 and 1, realizing multi-time period probability prediction of the four types of behavior. The multi-time period probability prediction includes predicting the probability of a user engaging in overdue payment behavior, policy consultation behavior, repair request behavior, and complaint tendency behavior in the next 1 day, 3 days, and 7 days.

[0029] In this embodiment, the static spatial attributes include: physical location attributes, pipeline affiliation attributes, and building attributes: the physical location attributes include the building, unit, floor, and latitude and longitude coordinates of the heat user; the pipeline affiliation attributes include the heat exchange station, heating pipeline, and heating station zone to which the heat user belongs; the building attributes include the building's age, insulation level, and apartment layout to which the heat user belongs. The regional association behaviors include: neighborhood association behaviors, pipeline topology association behaviors, and regional aggregation behaviors; the neighborhood association behaviors include repair requests, complaints, and overdue payment behaviors of other users in the same building, unit, and heat exchange station; the pipeline topology association behaviors include abnormal data on pressure, temperature, and flow of the pipeline where the user is located; the regional aggregation behaviors include the overdue payment rate, repair frequency, complaint density, and the supply and return water temperature, pressure, and energy consumption indicators of the corresponding heat exchange station in the area.

[0030] In this embodiment, the hot user payment ability label is based on the hot user's historical number of arrears, the proportion of arrears period, and the payment timeliness rate, and uses a weighted scoring method to quantify the hot user's willingness to pay. The room temperature sensitivity label is based on the frequency of room temperature repair requests, the frequency of room temperature inquiries, and the timeliness of feedback on room temperature fluctuations. A sensitivity scoring model is constructed using a gradient boosting tree to obtain the room temperature sensitivity score of heat users. The communication preference tags represent the user's preferences for communication channels, communication duration, communication style, and communication response. The complaint tendency label is quantified by a weighted scoring method based on the number of historical complaints, satisfaction with repair handling, percentage of negative emotions in conversations, percentage of unresolved issues, and complaint density in the same building.

[0031] In this embodiment, when a heat user only has overdue payment behavior in the future, the heating customer service intelligent agent is automatically triggered. Based on the heat user profile model, the agent analyzes the heat user's payment willingness score and communication preferences, and the digital human customer service calls upon a preset overdue payment knowledge base to match the heat user's payment collection strategy, conducting intelligent collection of overdue payments from the heat user, including: When the probability of a heating user's overdue payment behavior is greater than the preset threshold, and the probabilities of policy consultation behavior, repair request behavior, and complaint tendency behavior are all less than the preset threshold, it is determined that only overdue payment behavior has occurred, and the heating customer service intelligent agent is automatically triggered. The heating customer service intelligent agent dispatches the heat user profile model to obtain the heat user's payment willingness score and communication preference characteristics. It prioritizes matching the strategies in the preset overdue payment knowledge base according to the payment willingness level, core communication channels and payment ability tags. If a complete match cannot be made, it performs fuzzy adaptation according to the payment willingness level and outputs a collection instruction that includes communication channels, contact time, core strategies and script templates. The digital human customer service team intelligently reminds users with outstanding fees to pay according to the collection instructions. Furthermore, when a heat user only inquires about policies in a future time period, the heating customer service intelligent agent will be automatically triggered, and the digital human customer service will extract the hot keywords of the inquiry and automatically match the relevant policy content to respond, including: When the probability of a heating user inquiring about policies is greater than a preset threshold, and the probabilities of overdue payments, repair requests, and complaints are all less than the preset threshold, it is determined that only policy inquiries have occurred, and the heating customer service AI is automatically triggered. The heating customer service intelligent agent calls the digital human customer service to preprocess the policy consultation text of heating users. After extracting the hot words of policy consultation through word segmentation and keyword extraction algorithms, it calls the heating policy knowledge base, performs precise matching or semantic similarity fuzzy matching based on the hot words of consultation, obtains the corresponding policy content, and organizes it into a structured reply text to reply to heating users.

[0032] It should be noted that keyword extraction employs one of the following: TF-IDF, TextRank, or a pre-trained language model. This is used to extract high-frequency policy-related hot topics, such as deferred payment policies, suspension application procedures, and room temperature standards. Semantic similarity fuzzy matching calculates the vector similarity between hot topics and entries in the heating policy knowledge base, returning policy content with high similarity.

[0033] In this embodiment, when a heat user only requests repairs and / or has a tendency to complain in the future, the heating operation and maintenance intelligent agent is automatically triggered to conduct intelligent investigation of heating anomalies and dispatch work orders to operation and maintenance personnel. Simultaneously, the heating customer service intelligent agent is triggered to analyze the heat user's complaint tendency score and room temperature sensitivity score based on the heat user profile model. The digital human customer service representative then pushes personalized care messages based on the complaint tendency score, including: The heating operation and maintenance intelligent agent is triggered when the probability of a heat user reporting a repair request is greater than a preset threshold and the probability of other behaviors is less than a preset threshold; the heating operation and maintenance intelligent agent is triggered when the probability of a heat user complaining is greater than a preset threshold and the probability of other behaviors is less than a preset threshold; the heating operation and maintenance intelligent agent is triggered when the probability of both heat user reporting a repair request and complaining is greater than a preset threshold and the probability of other behaviors is less than a preset threshold. The heating operation and maintenance intelligent agent calls the built-in fault analysis model to analyze the operation data of the pipeline network and heat exchange station to which the target heat user belongs, as well as the heat metering data of the heat user's building or household. After analyzing the list of suspected fault triggering factors, it calls the built-in temperature difference analysis model to compare and analyze the temperature of the target heat user's community, building, and floors above and below, locate the root cause of the lack of heat, and generate a heating anomaly investigation report that includes the scope of the fault's impact and the root cause analysis report. The heating operation and maintenance intelligent agent intelligently matches the optimal operation and maintenance personnel based on the anomaly investigation report and the load of operation and maintenance personnel, and generates operation and maintenance work orders for dispatch. After the heating operation and maintenance intelligent agent investigates heating anomalies and dispatches operation and maintenance work orders, the heating customer service intelligent agent calls the heat user profile model to obtain the heat user complaint tendency score, heat user room temperature sensitivity score, and heat user communication preference. Based on the complaint tendency score and heat user room temperature sensitivity score, it classifies the users and calls the customer service care script knowledge base to match the corresponding care strategy. The digital human customer service pushes personalized care scripts according to the heat user communication preferences, and simultaneously informs the operation and maintenance work order dispatch status, the root cause of the fault or lack of heating, and the processing time.

[0034] It should be noted that during the anomaly investigation process, reports and complaints from heat users with high room temperature sensitivity are prioritized. During the investigation, the temperature data of the building and pipe network to which the heat user belongs will be closely monitored to expedite root cause identification and work order assignment. Additionally, heat users with a high complaint tendency will trigger pre-intervention by human customer service, while heat users with a medium to low complaint tendency will receive support and investigation progress updates from a digital customer service team.

[0035] In this embodiment, when a heating user defaults on their payment and exhibits a need for repairs or a tendency to complain, the heating operation and maintenance intelligent agent is automatically triggered to conduct intelligent troubleshooting of heating anomalies and dispatch work orders to maintenance personnel. Simultaneously, the heating customer service intelligent agent is triggered to analyze the user's payment willingness score, communication preference, complaint tendency score, and room temperature sensitivity score based on the user profile model. The digital human customer service then calls upon a pre-set overdue payment knowledge base to match the user's payment deferral strategy, prompting the user to apply for deferral and promoting personalized care messages, including: When the probability of a heating user's overdue payment is greater than a preset threshold, and the probability of a repair request or complaint is greater than a preset threshold, and the probability of a policy consultation is less than a preset threshold, the heating operation and maintenance intelligent agent is triggered. The heating operation and maintenance intelligent agent calls the built-in fault analysis model to analyze the operation data of the pipeline network and heat exchange station to which the target heat user belongs, as well as the heat metering data of the heat user's building or household. After analyzing the list of suspected fault triggering factors, it calls the built-in temperature difference analysis model to compare and analyze the temperature of the target heat user's community, building, and floors above and below, locate the root cause of the lack of heat, and generate a heating anomaly investigation report that includes the scope of the fault's impact and the root cause analysis report. The heating operation and maintenance intelligent agent intelligently matches the optimal operation and maintenance personnel based on the anomaly investigation report and the load of operation and maintenance personnel, and generates operation and maintenance work orders for dispatch. After the heating operation and maintenance intelligent agent investigates heating anomalies and dispatches operation and maintenance work orders, the heating customer service intelligent agent calls the heat user profile model to obtain the heat user's willingness to pay score, complaint tendency score, room temperature sensitivity score, and communication preference. Based on the willingness to pay classification, it matches the strategies in the preset overdue payment knowledge base and outputs the heat user's deferred payment strategy. At the same time, based on the complaint tendency score and the room temperature sensitivity score, it classifies and calls the customer service care script knowledge base to match the corresponding care strategy. The digital human customer service pushes the deferred payment application prompt and personalized care script according to the heat user's communication preference, and simultaneously informs the operation and maintenance work order dispatch status, the root cause of the fault or lack of heating, and the processing time.

[0036] It should be noted that, based on the payment willingness score, the expected payment knowledge base is accessed to match personalized deferred payment strategies, such as installment periods and late payment fee reduction rules. Combining the complaint tendency score and room temperature sensitivity score, the customer service care script knowledge base is accessed to match corresponding care strategies, including at least: High complaint tendency and high room temperature sensitivity: Emphasis is placed on prioritizing the handling of abnormal room temperature and supporting deferred payment policies; Complaint tendency and sensitivity to high room temperature: Inform about the progress of room temperature investigation and guidelines for deferred payment applications; Low complaint tendency: Pushing out explanations of deferred payment policies and timeliness of fault handling.

[0037] like Figure 5 As shown, in this embodiment, the step of training a human customer service intervention timing model based on historical dialogue data between the digital human customer service representative and popular users, popular user profile models, and human customer service intervention data to obtain the optimal timing for human customer service intervention includes: Acquire historical dialogue data between digital human customer service and popular users, popular user profile models, and human customer service intervention data as training sets; A human customer service intervention timing model is constructed, including a text feature extraction module, a profile feature fusion module, and an intervention timing prediction module. The training set is input into the human customer service intervention timing model, and the intervention accuracy and timing fit rate are used as training objectives to train the human customer service intervention timing model. During the interaction between the digital human customer service representative and the hot user, the dialogue data is collected in real time and the corresponding hot user profile data is retrieved. This data is then input into the trained human customer service intervention timing model, which outputs the probability of human customer service intervention and the optimal intervention time. When the intervention probability is greater than a preset threshold, the human customer service representative is automatically triggered to intervene in the dialogue at the optimal intervention time. The dialogue history and user profile information are also pushed to the human customer service representative for response.

[0038] It should be noted that the historical dialogue data between the digital human customer service and hot users includes: dialogue text, interaction duration, user response speed, dialogue emotion tags (such as anger, dissatisfaction, confusion), digital human's response results, and user interaction actions (such as refusing to communicate, asking complex questions), etc.; the human customer service intervention data includes: the timing of human customer service intervention (such as after the first interaction with the digital human, or when the user's emotions are negative), the reason for intervention, the handling results after intervention (such as complaint resolution, problem solving, user satisfaction), and the adverse consequences of not intervening in a timely manner (such as complaint escalation, user refusing to pay fees), etc. Intervention timing prediction: The model analyzes the input data in real time and outputs the intervention probability (0-1) and the optimal intervention time (such as after the second round of interaction with the digital human customer service, after the user's negative emotions have lasted for 30 seconds, or when the user asks a complex question).

[0039] like Figure 6 As shown, in this embodiment, the establishment of an optimized allocation model for heating customer service intervention, based on the constructed heterogeneous graph structure of the heating customer service system, includes: Based on historical processing data, business capability tags, performance scores and response time of heating customer service, a profile of heating customer service personnel is constructed, including professional capability score, communication style, load status and preferred heating user behavior scenarios. A heterogeneous graph of heating customer service is constructed with heat users, heating customer service representatives, and heat user behavior scenarios as nodes. At the same time, service association edges, capability matching edges, and load association edges are defined by combining heating customer service profiles, heat user profiles, and heat user behavior scenarios. The heterogeneous graph of heating human customer service is input into the heterogeneous graph attention network. The features of the heterogeneous graph are aggregated through node-level attention and semantic-level attention, and the feature vectors of heat user behavior demand and heating human customer service capability are output. Using the feature vectors of heat user behavior and the feature vectors of heating human customer service capabilities as input data, and with the objectives of maximizing matching degree, minimizing service cost, and optimizing response time, and constrained by the load, response time, and business capabilities of heating human customer service, an optimal allocation model for the intervention of heating human customer service is established to obtain the best heating human customer service.

[0040] In practical applications, service association edges represent the degree of association between heat users and the historical service interactions of heating customer service representatives; capability matching edges represent the degree of matching between the professional capabilities of heating customer service representatives and the current behavioral scenario needs of heat users; and load association edges represent the degree of fit between the current load status of heating customer service representatives and the urgency of heat user behavioral scenarios. Node-level attention calculation assigns weights to neighboring nodes of the same type, focusing on highly associated nodes; semantic-level attention calculation assigns weights to features of different meta-paths in the heterogeneous graph, focusing on core semantic paths. The heterogeneous graph attention network outputs a behavioral requirement feature vector for each heat user node, representing the heat user's own characteristics and associated behavioral scenario features, and a capability feature vector for each customer service representative node, representing the heating customer service representative's own characteristics and associated business capabilities and preferred service behavioral scenario features. Matching degree refers to the degree of matching between the heat user's behavioral needs and the capabilities of the customer service representative; service cost refers to the estimated time cost for the customer service representative to handle the heat user's situation; and response timeliness refers to the ratio of the actual response time to the required response time.

[0041] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0042] Furthermore, the functional modules in the various embodiments of this invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. If the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0043] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for integrated management and control of heating system production, billing, and customer service that incorporates AI, characterized in that: It includes: The system obtains heat user payment data, repair data, customer service data, and heat metering data from the heating billing subsystem, customer service subsystem, and production operation subsystem to form a heat user behavior sequence. After extracting time features, behavioral features, and spatial correlation features from the heat user behavior sequence, a spatiotemporal prediction model of heat user behavior trajectory is constructed to predict the behavior of heat users in different future time periods, including overdue payment behavior, policy consultation behavior, repair request behavior, and complaint tendency behavior. Construct a user profile model that includes multi-dimensional labels such as user payment ability, room temperature sensitivity, communication preferences, and complaint tendency; When a heating user only has overdue payment behavior in the future, the heating customer service intelligent agent will be automatically triggered. Based on the heating user profile model, the agent will analyze the heating user's willingness to pay and communication preferences. The digital human customer service will then call the preset overdue payment knowledge base, match the heating user's payment collection strategy, and carry out intelligent collection of overdue heating users. When a heating user only inquires about policies in the future, the heating customer service AI will be automatically triggered, and the digital human customer service will extract the hot keywords of the inquiry and automatically match the relevant policy content to reply. When a heat user only has a need for repair and / or a tendency to complain in the future, the heating operation and maintenance intelligent agent will be automatically triggered to conduct intelligent investigation of heating anomalies and dispatch work orders to operation and maintenance personnel. At the same time, the heating customer service intelligent agent will be triggered to analyze the heat user complaint tendency score and the heat user room temperature sensitivity score based on the heat user profile model, and the digital human customer service will push personalized care messages based on the complaint tendency score. When a heating user defaults on their payment and has a need for repairs or a tendency to complain, the heating operation and maintenance intelligent agent is automatically triggered to conduct intelligent investigation of heating anomalies and dispatch work orders to operation and maintenance personnel. At the same time, the heating customer service intelligent agent is triggered to analyze the heating user's payment willingness score, communication preference, complaint tendency score, and room temperature sensitivity score based on the heating user profile model. The digital human customer service calls up the preset overdue payment knowledge base, matches the heating user's fee deferral strategy, prompts the heating user to apply for deferral, and promotes personalized care messages. Based on the historical dialogue data between digital human customer service and heat users, the heat user profile model, and the human customer service intervention data, a human customer service intervention timing model is trained to obtain the optimal timing for human customer service intervention. Then, combined with the constructed heterogeneous graph structure of heating human customer service, an optimized allocation model for heating human customer service intervention is established to allocate the optimal heating human customer service for dialogue services.

2. The integrated management and control method for production, billing, and customer service of a heating system according to claim 1, characterized in that, The data obtained from the heating billing subsystem includes: basic information of the heating user, heating payment time, payment amount, arrears duration, payment method, arrears frequency, and tiered heating consumption. The data obtained from the customer service subsystem includes: repair report data, fault type, fault location, and processing time; the time, type, content, processing result, and heating user sentiment of customer service inquiries and complaints. The data obtained from the production and operation subsystem includes: supply and return water temperature and pressure of the heat exchange station, the heat exchange station to which the heating user belongs, pipeline zones, heat leakage rate of the area to which the heating user belongs, and fault occurrence rate. The formation of the heat user behavior sequence includes: using the heat user's unique account number as an identifier and using time period as the granularity, forming a time-series and structured heat user behavior sequence from heat payment data, repair data, customer service data, and heat metering data.

3. The integrated management and control method for production, billing, and customer service of a heating system according to claim 1, characterized in that, The extraction of time features from the hot user behavior sequence includes: using a short-term local time-series feature extraction module to extract short-term local time feature vectors of hot user behavior; using a long-term time-series dependency learning module to process the short-term local time feature vectors, learn the long-term time-series dependencies of hot user behavior, and output long-term time feature vectors; and using a dynamic weight fusion module to adaptively allocate weights to the short-term local time features and long-term time feature vectors, and fuse them to obtain a time feature vector. The extraction of behavioral features from the hot user behavior sequence includes: using an attention-enhanced feedforward neural network to perform nonlinear mapping on the hot user behavior sequence, mining the influence of each data in the behavior sequence on the hot user behavior trajectory, as well as the coupling relationship between each behavior, and outputting a behavioral feature vector; The spatial correlation feature extraction of the heat user behavior sequence includes: using heat users, heat exchange stations, and pipeline zones as graph nodes, and using the subordinate and adjacent relationships between heat users and heat exchange stations, users in the same pipeline zone, and heat exchange stations and associated pipelines as edges to construct a heating spatial topology graph; generating an adjacency matrix based on the heating spatial topology graph and performing normalization processing to eliminate the feature value deviation between the central node and the edge node to obtain a normalized adjacency matrix; using the heating spatial topology graph and the normalized adjacency matrix as input, using a graph convolutional neural network (GCN) to aggregate spatial features of the static spatial attributes and regional correlation behaviors of heat users, learning local spatial correlation and global spatial topology, and outputting an initial spatial feature vector; and performing masking optimization on the initial spatial feature vector based on the subordinate relationships of the heating spatial topology graph to mask the relevant features of non-associated heat exchange stations and pipeline zones, retaining key spatial features, and obtaining the final spatial correlation feature vector. Furthermore, the construction of a spatiotemporal prediction model for heat user behavior trajectories, predicting the behavior of heat users at different future time periods, includes: fusing temporal features, behavioral features, and spatial correlation features on a single user and its local space as units, outputting local spatiotemporal behavioral features; performing probability distribution learning on the local spatiotemporal behavioral features to mine the spatiotemporal behavioral correlation within the global heating supply scope, outputting global spatiotemporal behavioral features; using the global spatiotemporal behavioral features as input, employing a network structure with a fully connected layer and a multi-task output layer, performing feature learning and model training, constructing a spatiotemporal prediction model for heat user behavior trajectories, outputting the probability values ​​of a single heat user's overdue payment behavior, policy consultation behavior, repair request behavior, and complaint tendency behavior at different future time periods, and simultaneously outputting regional-level heat user behavior prediction data; wherein, the repair request behavior refers to heat user repair requests due to equipment or pipeline failures or substandard room temperature; the complaint tendency behavior refers to heat user complaints due to substandard room temperature or unsatisfactory fault handling.

4. The integrated management and control method for production, billing, and customer service of a heating system according to claim 3, characterized in that, The static spatial attributes include: physical location attributes, pipeline affiliation attributes, and building attributes. The physical location attributes include the building, unit, floor, and latitude and longitude coordinates of the heat user. The pipeline affiliation attributes include the heat exchange station, heating pipeline, and heating station zone to which the heat user belongs. The building attributes include the building's age, insulation level, and apartment layout. The regional association behaviors include: neighborhood association behaviors, pipeline topology association behaviors, and regional aggregation behaviors; the neighborhood association behaviors include repair requests, complaints, and overdue payment behaviors of other users in the same building, unit, and heat exchange station; the pipeline topology association behaviors include abnormal data on pressure, temperature, and flow of the pipeline where the user is located; the regional aggregation behaviors include the overdue payment rate, repair frequency, complaint density, and the supply and return water temperature, pressure, and energy consumption indicators of the corresponding heat exchange station in the area.

5. The integrated management and control method for production, billing, and customer service of a heating system according to claim 1, characterized in that, The hot user payment ability label is based on the hot user's historical number of arrears, the proportion of arrears period, and the payment timeliness rate, and uses a weighted scoring method to quantify the hot user's willingness to pay. The room temperature sensitivity label is based on the frequency of room temperature repair requests, the frequency of room temperature inquiries, and the timeliness of feedback on room temperature fluctuations. A sensitivity scoring model is constructed using a gradient boosting tree to obtain the room temperature sensitivity score of heat users. The communication preference tags represent the user's preferences for communication channels, communication duration, communication style, and communication response. The complaint tendency label is quantified by a weighted scoring method based on the number of historical complaints, satisfaction with repair handling, percentage of negative emotions in conversations, percentage of unresolved issues, and complaint density in the same building.

6. The integrated management and control method for production, billing, and customer service of a heating system according to claim 1, characterized in that, When a heating user only has overdue payment behavior in the future, the heating customer service intelligent agent is automatically triggered. Based on the heating user profile model, it analyzes the user's willingness to pay and communication preferences, and the digital human customer service calls upon a preset overdue payment knowledge base to match the user's payment collection strategy, conducting intelligent collection of overdue payments, including: When the probability of a heating user's overdue payment behavior is greater than the preset threshold, and the probabilities of policy consultation behavior, repair request behavior, and complaint tendency behavior are all less than the preset threshold, it is determined that only overdue payment behavior has occurred, and the heating customer service intelligent agent is automatically triggered. The heating customer service intelligent agent dispatches the heat user profile model to obtain the heat user's payment willingness score and communication preference characteristics. It prioritizes matching the strategies in the preset overdue payment knowledge base according to the payment willingness level, core communication channels and payment ability tags. If a complete match cannot be made, it performs fuzzy adaptation according to the payment willingness level and outputs a collection instruction that includes communication channels, contact time, core strategies and script templates. The digital human customer service team intelligently reminds users with outstanding fees to pay according to the collection instructions. Furthermore, when a heat user only inquires about policies in a future time period, the heating customer service intelligent agent will be automatically triggered, and the digital human customer service will extract the hot keywords of the inquiry and automatically match the relevant policy content to respond, including: When the probability of a heating user inquiring about policies is greater than a preset threshold, and the probabilities of overdue payments, repair requests, and complaints are all less than the preset threshold, it is determined that only policy inquiries have occurred, and the heating customer service AI is automatically triggered. The heating customer service intelligent agent calls the digital human customer service to preprocess the policy consultation text of heating users. After extracting the hot words of policy consultation through word segmentation and keyword extraction algorithms, it calls the heating policy knowledge base, performs precise matching or semantic similarity fuzzy matching based on the hot words of consultation, obtains the corresponding policy content, and organizes it into a structured reply text to reply to heating users.

7. The integrated management and control method for production, billing, and customer service of a heating system according to claim 1, characterized in that, When a heat user only requests repairs and / or has a tendency to complain in the future, the heating operation and maintenance intelligent agent is automatically triggered to conduct intelligent investigation of heating anomalies and dispatch work orders to operation and maintenance personnel. Simultaneously, the heating customer service intelligent agent is triggered to analyze the heat user's complaint tendency score and room temperature sensitivity score based on the heat user profile model. The digital human customer service representative then pushes personalized care messages based on the complaint tendency score, including: The heating operation and maintenance intelligent agent is triggered when the probability of a heat user reporting a repair request is greater than a preset threshold and the probability of other behaviors is less than a preset threshold; the heating operation and maintenance intelligent agent is triggered when the probability of a heat user complaining is greater than a preset threshold and the probability of other behaviors is less than a preset threshold; the heating operation and maintenance intelligent agent is triggered when the probability of both heat user reporting a repair request and complaining is greater than a preset threshold and the probability of other behaviors is less than a preset threshold. The heating operation and maintenance intelligent agent calls the built-in fault analysis model to analyze the operation data of the pipeline network and heat exchange station to which the target heat user belongs, as well as the heat metering data of the heat user's building or household. After analyzing the list of suspected fault triggering factors, it calls the built-in temperature difference analysis model to compare and analyze the temperature of the target heat user's community, building, and floors above and below, locate the root cause of the lack of heat, and generate a heating anomaly investigation report that includes the scope of the fault's impact and the root cause analysis report. The heating operation and maintenance intelligent agent intelligently matches the optimal operation and maintenance personnel based on the anomaly investigation report and the load of operation and maintenance personnel, and generates operation and maintenance work orders for dispatch. After the heating operation and maintenance intelligent agent investigates heating anomalies and dispatches operation and maintenance work orders, the heating customer service intelligent agent calls the heat user profile model to obtain the heat user complaint tendency score, heat user room temperature sensitivity score, and heat user communication preference. Based on the complaint tendency score and heat user room temperature sensitivity score, it classifies the users and calls the customer service care script knowledge base to match the corresponding care strategy. The digital human customer service pushes personalized care scripts according to the heat user communication preferences, and simultaneously informs the operation and maintenance work order dispatch status, the root cause of the fault or lack of heating, and the processing time.

8. The integrated management and control method for production, billing, and customer service of a heating system according to claim 1, characterized in that, When a heating user defaults on their payment and exhibits signs of needing repairs or a tendency to complain, the system automatically triggers the heating operation and maintenance intelligent agent to conduct intelligent troubleshooting of heating anomalies and dispatch work orders to maintenance personnel. Simultaneously, it triggers the heating customer service intelligent agent, which analyzes the user's payment willingness score, communication preferences, complaint tendency score, and room temperature sensitivity score based on the user profile model. The digital customer service representative then accesses a pre-set overdue payment knowledge base, matches the user with a payment deferral strategy, prompts the user to apply for deferral, and promotes personalized customer care messages, including: When the probability of a heating user's overdue payment is greater than a preset threshold, and the probability of a repair request or complaint is greater than a preset threshold, and the probability of a policy consultation is less than a preset threshold, the heating operation and maintenance intelligent agent is triggered. The heating operation and maintenance intelligent agent calls the built-in fault analysis model to analyze the operation data of the pipeline network and heat exchange station to which the target heat user belongs, as well as the heat metering data of the heat user's building or household. After analyzing the list of suspected fault triggering factors, it calls the built-in temperature difference analysis model to compare and analyze the temperature of the target heat user's community, building, and floors above and below, locate the root cause of the lack of heat, and generate a heating anomaly investigation report that includes the scope of the fault's impact and the root cause analysis report. The heating operation and maintenance intelligent agent intelligently matches the optimal operation and maintenance personnel based on the anomaly investigation report and the load of operation and maintenance personnel, and generates operation and maintenance work orders for dispatch. After the heating operation and maintenance intelligent agent investigates heating anomalies and dispatches operation and maintenance work orders, the heating customer service intelligent agent calls the heat user profile model to obtain the heat user's willingness to pay score, complaint tendency score, room temperature sensitivity score, and communication preference. Based on the willingness to pay classification, it matches the strategies in the preset overdue payment knowledge base and outputs the heat user's deferred payment strategy. At the same time, based on the complaint tendency score and the room temperature sensitivity score, it classifies and calls the customer service care script knowledge base to match the corresponding care strategy. The digital human customer service pushes the deferred payment application prompt and personalized care script according to the heat user's communication preference, and simultaneously informs the operation and maintenance work order dispatch status, the root cause of the fault or lack of heating, and the processing time.

9. The integrated management and control method for production, billing, and customer service of a heating system according to claim 1, characterized in that, The process involves training a human customer service intervention timing model based on historical dialogue data between digital human customer service representatives and popular users, popular user profile models, and human customer service intervention data to obtain the optimal timing for human customer service intervention in the dialogue, including: Acquire historical dialogue data between digital human customer service and popular users, popular user profile models, and human customer service intervention data as training sets; A human customer service intervention timing model is constructed, including a text feature extraction module, a profile feature fusion module, and an intervention timing prediction module. The training set is input into the human customer service intervention timing model, and the intervention accuracy and timing fit rate are used as training objectives to train the human customer service intervention timing model. During the interaction between the digital human customer service representative and the hot user, the dialogue data is collected in real time and the corresponding hot user profile data is retrieved. This data is then input into the trained human customer service intervention timing model, which outputs the probability of human customer service intervention and the optimal intervention time. When the intervention probability is greater than a preset threshold, the human customer service representative is automatically triggered to intervene in the dialogue at the optimal intervention time. The dialogue history and user profile information are also pushed to the human customer service representative for response.

10. The integrated management and control method for production, billing, and customer service of a heating system according to claim 8, characterized in that, The aforementioned combination of the constructed heterogeneous graph structure of heating customer service and the establishment of an optimized allocation model for heating customer service intervention includes: Based on historical processing data, business capability tags, performance scores and response time of heating customer service, a profile of heating customer service personnel is constructed, including professional capability score, communication style, load status and preferred heating user behavior scenarios. A heterogeneous graph of heating customer service is constructed with heat users, heating customer service representatives, and heat user behavior scenarios as nodes. At the same time, service association edges, capability matching edges, and load association edges are defined by combining heating customer service profiles, heat user profiles, and heat user behavior scenarios. The heterogeneous graph of heating human customer service is input into the heterogeneous graph attention network. The features of the heterogeneous graph are aggregated through node-level attention and semantic-level attention, and the feature vectors of heat user behavior demand and heating human customer service capability are output. Using the feature vectors of heat user behavior and the feature vectors of heating human customer service capabilities as input data, and with the objectives of maximizing matching degree, minimizing service cost, and optimizing response time, and constrained by the load, response time, and business capabilities of heating human customer service, an optimal allocation model for the intervention of heating human customer service is established to obtain the best heating human customer service.

Images