A large model-based home guest network industry collaborative planning and decision-making method

Abstract

The application particularly relates to a large model-based home guest network industry collaborative planning and decision method. The large model-based home guest network industry collaborative planning and decision method collects network data, business data and customer data of a home guest network, integrates the data into a unified data warehouse after preprocessing, and establishes an association relationship; demand and target of planning and decision are self-defined, current real-time data are input into a large model after preprocessing for inference calculation to generate a plurality of candidate schemes; an optimal scheme is determined through comprehensive evaluation, the scheme is decomposed into specific implementation tasks, an implementation plan is formulated, and resources are allocated for implementation. The large model-based home guest network industry collaborative planning and decision method realizes accurate matching of network resources and business development, improves operation efficiency and service quality of the home guest network, reduces operation cost of the enterprise, enhances core competitiveness of the enterprise in the market, and provides more high-quality and more personalized network service and business experience for home guest users.

Description

Technical Field

[0001] This invention relates to the field of communication network technology, and in particular to a collaborative planning and decision-making method for home and customer networks based on a large model. Background Technology

[0002] In recent years, the global communications industry has undergone unprecedented changes, and home networks, as an important part of communication networks, have seen their scale and complexity continuously increase.

[0003] According to relevant industry reports, by the end of 2024, the penetration rate of fixed broadband in Chinese households had exceeded 90%, and the number of 5G household users surpassed 300 million. In terms of services, high-definition video and online gaming were the two services with the highest traffic share among home users, accounting for 45% and 20% of total traffic, respectively. Furthermore, with the normalization of remote work and online education, the user base for these two services has maintained an average annual growth rate of over 15%.

[0004] Currently, operators mainly use the traditional "offline + manual + experience" approach for planning and decision-making of home customer networks. This approach suffers from problems such as separation of network planning and business decision-making, insufficient basic data processing, poor dynamic adaptability of planning and decision-making, and lack of a quantitative evaluation system. As a result, home customer network planning is unable to meet the needs of home customer business development.

[0005] (1) Separation of network planning and business decision-making Network planning is primarily the responsibility of the network department, with its core objective being to ensure network stability, reliability, and scalability. The planning process often relies heavily on historical network performance data, lacking forward-looking consideration of business development trends. Business decisions are dominated by the marketing or business departments, focusing more on market demands, user preferences, and competitor activities, with insufficient consideration for the network's actual carrying capacity.

[0006] Insufficient planning and business coordination, often relying on manual decision-making, leads to one-sidedness in customer network planning, making it difficult to balance network and business needs, ultimately resulting in poor network planning effectiveness and an inability to efficiently support market business development needs.

[0007] (2) Insufficient basic data processing capabilities The data involved in the home customer network and services is huge and complex, including structured data (such as basic user information, network equipment parameters, service activation records, etc.), semi-structured data (such as user complaint texts, network logs, etc.), and unstructured data (such as user online behavior trajectories, video content preferences, etc.).

[0008] Traditional offline data collection and manual processing methods can only collect small amounts of data and perform simple statistical analysis. They have significant limitations when dealing with large-scale networks and massive amounts of heterogeneous data.

[0009] (3) Poor dynamic adaptability of planning and decision-making The home customer market environment and user needs are constantly changing. For example, the emergence of new technologies (such as the research and application of 6G), policy adjustments (such as updates to network security regulations), and changes in competitors' strategies (such as the launch of more cost-effective service packages) can all have a significant impact on home customer network planning and business decisions.

[0010] Most existing planning and decision-making methods are static. Once a planning scheme or business strategy is formulated, it is difficult to make quick adjustments based on the actual situation during implementation.

[0011] (4) Lack of a quantitative evaluation system The lack of a scientific and comprehensive quantitative evaluation system in the current planning and decision-making of the home customer network makes it difficult to measure the effectiveness of planning schemes and business strategies.

[0012] This lack of quantitative assessment makes it difficult for companies to summarize lessons learned and continuously optimize planning and decision-making methods, resulting in a long-term low level of planning and decision-making quality.

[0013] To overcome the problems of existing technologies, such as the separation of home and customer network planning and business decision-making, insufficient data processing capabilities, poor dynamic adaptability, and lack of quantitative evaluation system, this invention proposes a home and customer network collaborative planning and decision-making method based on a large model. Summary of the Invention

[0014] To overcome the shortcomings of existing technologies, this invention provides a simple and efficient collaborative planning and decision-making method for home and customer networks based on a large model.

[0015] This invention is achieved through the following technical solution: A collaborative planning and decision-making method for home and customer networks based on a large model includes the following steps: Step S1: Data Collection and Processing Collect network data, business data, and customer data from the home customer network, and clean the collected data to remove duplicate, erroneous, and incomplete data records; Data standardization methods are used to normalize data of different magnitudes and units; missing data is filled in; and the processed data is integrated into a unified data warehouse to establish relationships. In step S1, the network data of the home customer network includes topology, performance data and equipment configuration data; the service data includes the usage of home customer services, complaint data and competition data; and the customer data includes basic customer information and internet access behavior data.

[0016] Step S2, Home Customer Network Industry Collaborative Planning and Decision-Making Process Customize the planning and decision-making needs and objectives, and set quantitative objectives that conform to the SMART principle; preprocess the current real-time network data, real-time business data, and real-time customer data, and then input them into the trained large model; The large model performs inference calculations on the input data to generate several candidate home-customer network collaborative planning and decision-making schemes; the analytic hierarchy process is used to comprehensively evaluate the candidate schemes based on evaluation indicators to determine the optimal scheme; Break down the optimal solution into specific implementation tasks, formulate an implementation plan, and allocate resources for implementation; During implementation, task progress, network performance changes, business operations, and customer feedback are monitored in real time, and the feedback data is input into the large model. Based on the model evaluation results and actual monitoring, the solution is adjusted and iterated, and the relevant data is fed back into the data warehouse and the model training process.

[0017] In step S2, the collaborative planning and decision-making process of Jiake.com is as follows: Step S2.1: Requirements Analysis and Goal Setting Before making collaborative planning decisions, collect the needs and expectations of each department and set clear and quantifiable planning and decision-making goals; Step S2.2, Data Input and Preprocessing Preprocess the real-time data input to the large model, including data cleaning, standardization, and integration, to ensure that the quality and format of the input data are consistent with the requirements of the large model; The current real-time data is input into the trained large model, including real-time network data, real-time business data, and real-time customer data; Step S2.3, Model Reasoning and Solution Generation The preprocessed real-time data and the set planning and decision-making objectives are input into the trained large model, which then performs inference and calculation to generate several candidate home-customer network collaborative planning and decision-making schemes. During the inference process, the model comprehensively considers network status, business needs, customer preferences and market competition factors in real-time data, as well as the set quantitative targets, to conduct multi-dimensional analysis and prediction. The number of candidate solutions generated is 5-10, and each solution contains detailed network planning and business decision-making content; Step S2.4, Scheme Evaluation and Optimization Establish a comprehensive evaluation index system to evaluate candidate solutions from multiple dimensions. The evaluation indicators include economic indicators, technical indicators, market indicators, and customer indicators. The Analytic Hierarchy Process (AHP) is used to comprehensively evaluate candidate solutions. In step S2.4, the candidate solution evaluation method is as follows: First, an evaluation team composed of network experts, business experts, market experts, and customer representatives is invited to score the importance of each evaluation indicator and determine the indicator weights. Then, based on the actual or predicted data of each indicator, the performance of each candidate solution on each indicator is scored. Finally, based on the indicator weights and solution scores, the comprehensive score of each candidate solution is calculated, and the solution with the highest comprehensive score is the preliminary optimal solution.

[0018] For the preliminary optimal solution, relevant personnel are organized to discuss and review it, identify problems or deficiencies in the solution, and propose optimization suggestions; based on the discussion results, the solution is modified and improved to form the final optimal home-customer network collaborative planning and decision-making solution.

[0019] Step S2.5: Implementation and Monitoring of the Solution The final planning and decision-making scheme is broken down into specific implementation tasks, clearly defining the responsible departments, persons in charge, implementation timelines, and required resources for each task; and human, material, and financial resources are rationally allocated according to the implementation plan. In step S2.5, a real-time monitoring mechanism is established during the implementation of the solution. Through data acquisition equipment and business systems, the implementation progress of various tasks and the implementation effect of the solution are tracked in real time. The monitoring content includes task progress, network performance changes, business operation status and customer feedback. Real-time monitoring data is fed back to the large model in a timely manner. The model re-evaluates and predicts the implementation effect of the plan based on the new data. If the deviation between the actual effect and the expected effect exceeds a custom threshold, the model generates suggestions for adjusting the plan.

[0020] Step S2.6, Scheme Adjustment and Iteration Based on the adjustment suggestions generated by the model and the actual monitoring situation, the implemented plan is adjusted in a timely manner; After the implementation of the plan, a comprehensive summary and evaluation of the overall effect will be conducted, analyzing the successful experiences and shortcomings. Relevant data and experience will be fed back into the data warehouse and model training process to optimize the performance of the large model and improve the future collaborative planning and decision-making process of the home customer network, forming a virtuous cycle of continuous iteration.

[0021] In step S2, based on the Transformer architecture, customized improvements are made to meet the needs and objectives of collaborative planning and decision-making in the home and customer network industry, and a large model is constructed. The integrated historical data in the data warehouse is used as the original dataset for model training. The dataset is divided into training set, validation set and test set in chronological order, with a ratio of 7:1:2. The training and validation sets are labeled with labels including: network planning labels, business decision labels, and effect evaluation labels. Data annotation employs a combination of manual and automatic annotation methods: For historical data that already contains clear planning and decision-making schemes and implementation effect records, an automatic labeling method is used to directly extract relevant information as tags; For cases where there are no clear records or where further optimization is needed, a labeling team composed of network planning experts, business analysts, and data scientists performs manual labeling, and cross-validation by multiple people ensures the accuracy of the labeling. The large model is trained, validated, and optimized; the model training process is as follows: (1) Initialize parameters The initial parameters of the large model are initialized using the parameters of the pre-trained model. The pre-trained model is a Transformer model trained on large-scale general text corpora and communication industry-related data. For the custom layers in the model, the parameters are initialized using the Xavier initialization method. (2) Set training parameters Learning rate: The initial learning rate is set to 5e-5, and a cosine annealing learning rate scheduling strategy is adopted. The learning rate is gradually reduced as the number of training rounds increases. Batch size: Set to 32 based on the memory capacity of the training device, meaning that 32 data points are input for training at the same time each time; Training rounds: The total number of training rounds is 100. If the performance on the validation set does not improve for 10 consecutive rounds, training will be stopped early to prevent the model from overfitting.

[0022] Loss function: The cross-entropy loss function is used to calculate the loss between the model prediction result and the labeled label. For regression tasks, the mean squared error loss function is used. The total loss is the weighted sum of the classification loss and the regression loss, with a classification loss weight of 0.6 and a regression loss weight of 0.4. (3) Model training and monitoring Parallel training is performed using a distributed training framework. During training, after each round of training, the large model is evaluated using a validation set. The prediction accuracy, mean squared error, and other metrics of the large model are calculated, and the model parameters are recorded. At the same time, the changes in the model's loss on the training and validation sets are monitored. If the loss on the training set continues to decrease while the loss on the validation set increases, it indicates that the large model is overfitting. In this case, an early stopping strategy is adopted to stop training, and the model parameters with the best performance on the validation set are loaded. The model evaluation and optimization process is as follows: (1) Setting evaluation indicators The following metrics were used to evaluate the model performance: accuracy, root mean square error (RMSE), and decision performance metrics.

[0023] (2) Model optimization Based on the evaluation results, the large model was optimized in a targeted manner: If the model's evaluation metric on a certain type of data is lower than a custom threshold, then increase the proportion of that type of data in the training set and perform targeted fine-tuning training. If the large model exhibits overfitting, dropout regularization is employed by adding dropout layers to the encoding and decoding layers with a dropout rate of 0.1 to reduce excessive dependence between neurons. If the convergence speed of a large model is slow, adjust the learning rate scheduling strategy or increase the batch size to speed up the training process of the model. If the prediction accuracy of the large model does not meet the user's requirements, the depth and width of the large model are increased to improve the model's feature extraction and representation capabilities. After multiple rounds of evaluation and optimization, the final large model achieved an accuracy of over 90% on the test set, with the root mean square error (RMSE) controlled within 5% and the deviation of the decision performance index less than 10%. Training was then stopped, and the trained large model was obtained.

[0024] A large-scale model-based collaborative planning and decision-making system for home and customer networks, used to implement the above method, includes: Data collection module: Used to collect network data, business data, and customer data, including network data collection submodule, business data collection submodule, and customer data collection submodule; Among them, the network data acquisition submodule is responsible for collecting topology and performance data through the network management system and monitoring equipment, the business data acquisition submodule is responsible for obtaining business data from the business support system and complaint management system, and the customer data acquisition submodule is responsible for integrating customer information in the customer relationship management system. Data processing module: Used to process the collected data, including data cleaning unit, standardization unit, missing value imputation unit and data integration unit; Among them, the data cleaning unit is responsible for removing duplicate and erroneous data, the standardization unit is responsible for normalizing and encoding the data, the missing value filling unit is responsible for filling missing data using multiple methods, and the data integration unit is responsible for integrating the processed data into the data warehouse. Large Model Training Module: Used to build, train, and optimize large models, including model architecture design unit, training data preparation unit, parameter setting unit, model training unit, and model evaluation and optimization unit; The model architecture design unit is responsible for designing a large model structure based on Transformer; the training data preparation unit is responsible for building and partitioning the dataset and labeling it; the parameter setting unit is responsible for setting the training parameters; the model training unit is responsible for training the model using the training set; and the model evaluation and optimization unit is responsible for evaluating the model performance and optimizing it.

[0025] Collaborative Planning and Decision-Making Module: Used to execute the collaborative planning and decision-making process of the home customer network industry, including a demand target setting unit, a data input preprocessing unit, a solution generation unit, a solution evaluation and optimization unit, a solution implementation monitoring unit, and a solution adjustment and iteration unit; The unit is responsible for collecting requirements and setting targets; the data input preprocessing unit is responsible for inputting and preprocessing real-time data; the solution generation unit is responsible for generating candidate solutions using a large model; the solution evaluation and optimization unit is responsible for evaluating and optimizing solutions; the solution implementation monitoring unit is responsible for developing plans and monitoring implementation; and the solution adjustment and iteration unit is responsible for adjusting solutions and providing data feedback.

[0026] A home-and-guest network collaborative planning and decision-making device based on a large model includes a memory and a processor; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-described method steps.

[0027] A readable storage medium storing a computer program that, when executed by a processor, implements the above-described method steps.

[0028] The beneficial effects of this invention are as follows: This large-scale model-based collaborative planning and decision-making method for home and customer networks breaks down the barriers between network planning and business decision-making, constructs a collaborative mechanism between the two, enables network planning to fully consider business needs, and enables business decisions to be based on the actual network capabilities, achieving precise matching between network resources and business development; it improves the operational efficiency and service quality of home and customer networks, reduces the operating costs of enterprises, enhances the core competitiveness of enterprises in the market, and provides home and customer users with higher quality and more personalized network services and business experiences. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the 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 based on these drawings without creative effort.

[0030] Figure 1 This is a schematic diagram of the large model architecture based on the Transformer architecture of this invention.

[0031] Figure 2 This is a schematic diagram of the collaborative planning and decision-making process for the home customer network industry in this invention.

[0032] Figure 3 This is a schematic diagram of the model training and optimization process of the present invention. Detailed Implementation

[0033] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions in the embodiments of this invention will be clearly and completely described below in conjunction with the embodiments of this invention. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0034] This large-scale model-based collaborative planning and decision-making method for the home and customer network includes the following steps: Step S1: Data Collection and Processing Collect network data, business data, and customer data from the home customer network, and clean the collected data to remove duplicate, erroneous, and incomplete data records; Data standardization methods are used to normalize data of different magnitudes and units; missing data is filled in; and the processed data is integrated into a unified data warehouse to establish relationships. In step S1, the network data of the home customer network includes topology, performance data and equipment configuration data; the service data includes the usage of home customer services, complaint data and competition data; and the customer data includes basic customer information and internet access behavior data.

[0035] The network data collection process of Jiake Network is as follows: (1) Network topology data The network management system uses a topology discovery tool to collect real-time physical and logical topology data of the home and guest networks. Physical topology data includes fiber optic cable laying paths, cable types and lengths, optical splitter locations and connections, base station locations and coverage areas, and user-end equipment (such as optical modems and routers) installation locations and device identifiers.

[0036] (2) Network performance data By utilizing monitoring devices distributed across various network nodes, such as network probes and traffic analyzers, network performance metrics are collected in real time. These metrics include: bandwidth utilization, latency, packet loss rate, jitter, and availability.

[0037] (3) Network device configuration information The network configuration management system collects detailed configuration parameters for all network devices, including routing protocol configurations, ACL (Access Control List) rules, and QoS (Quality of Service) policies for routers; port configurations, VLAN configurations, and link aggregation configurations for switches; and speed configurations and authentication methods for optical modems. It also records device firmware versions, hardware models, and production dates to provide a basis for device upgrades, maintenance, and replacements. Device configuration information is collected through automatic backups, with a backup of all device configurations performed every morning at midnight and compared with the previous backup, recording any configuration changes.

[0038] The business data of the home customer network reflects the operational status and market demand of home customer services and is an important basis for business decision-making. The business data collection process is as follows: (1) Business usage data Extract usage data for various customer services from the business support system, including: service activation and cancellation, service usage duration, service traffic consumption, and service access frequency.

[0039] (2) Customer complaint data Establish a unified customer complaint management system to collect complaint information submitted by users through channels such as customer service hotlines, app feedback, and social media. Complaint data includes: complainant information, complaint content, complaint time, and complaint processing status.

[0040] (3) Market competition data Market competition data is collected through market research, purchasing data from third-party data agencies, and analyzing publicly available competitor information. This includes competitor service packages, network coverage and service quality, and changes in market share.

[0041] Customer data is crucial for understanding user needs and providing personalized services, and therefore requires comprehensive and accurate collection and management. The customer data collection process is as follows: (1) Customer basic information Obtain basic customer information from the customer relationship management (CRM) system, including personal information, family information, and consumption information.

[0042] (2) Customer Internet behavior data We collect customer internet behavior data through methods such as network log analysis and business platform access records, including: internet access time, accessed content, and device usage.

[0043] In step S1, the data processing and preprocessing process is as follows: (1) Data cleaning The collected raw data is processed by combining automated data cleaning tools with manual review, including removing duplicate data, correcting erroneous data, and processing incomplete data.

[0044] (2) Data standardization To enable unified analysis of data of different types and scales, data standardization processes are performed, including the transformation of numerical data, categorical data, and text data.

[0045] (3) Data integration The cleaned and standardized network data, business data, and customer data are integrated into a unified data warehouse, and relationships are established.

[0046] Step S2, Home Customer Network Industry Collaborative Planning and Decision-Making Process Customize the planning and decision-making needs and objectives, and set quantitative objectives that conform to the SMART principle; preprocess the current real-time network data, real-time business data, and real-time customer data, and then input them into the trained large model; The large model performs inference calculations on the input data to generate several candidate home-customer network collaborative planning and decision-making schemes; the analytic hierarchy process is used to comprehensively evaluate the candidate schemes based on evaluation indicators to determine the optimal scheme; Break down the optimal solution into specific implementation tasks, formulate an implementation plan, and allocate resources for implementation; During implementation, task progress, network performance changes, business operations, and customer feedback are monitored in real time, and the feedback data is input into the large model. Based on the model evaluation results and actual monitoring, the solution is adjusted and iterated, and the relevant data is fed back into the data warehouse and the model training process.

[0047] In step S2, the collaborative planning and decision-making process of Jiake.com is as follows: Step S2.1: Requirements Analysis and Goal Setting Before making collaborative planning decisions, it is essential to first clarify the needs and objectives of the planning decisions. This involves communicating with relevant personnel in the company's network, business, and marketing departments to gather their needs and expectations. For example, the network department may want to reduce network failure rates and improve network stability through planning; the business department may want to increase business revenue and market share through decisions; and the marketing department may want to improve customer satisfaction and enhance brand influence.

[0048] Based on the needs of each department, set clear and quantifiable planning and decision-making goals. Goals should conform to the SMART principle: Specific, Measurable, Achievable, Relevant, and Time-bound. For example, goals could include reducing network failure rate by 20%, increasing business revenue by 15%, and raising customer satisfaction to over 90 points within the next six months.

[0049] Step S2.2, Data Input and Preprocessing Perform the same preprocessing operations on the real-time data input to the large model as on the training data, including data cleaning, standardization, and integration, to ensure that the quality and format of the input data are consistent with the requirements of the large model.

[0050] The current real-time data is input into the trained large model, including: real-time network data, real-time business data, and real-time customer data; Step S2.3, Model Reasoning and Solution Generation The preprocessed real-time data and the set planning and decision-making objectives are input into the trained large model, which then performs inference and calculation to generate several candidate home-customer network collaborative planning and decision-making schemes. During the inference process, the model comprehensively considers network status, business needs, customer preferences, and market competition factors in real-time data, as well as set quantitative targets, to conduct multi-dimensional analysis and prediction. For example, the model analyzes the current network bottlenecks and predicts the improvement effects of different network planning measures on network performance; it also analyzes the impact of different business decision-making strategies on business revenue and customer satisfaction, as well as the feasibility of these decision-making strategies under current network conditions.

[0051] The number of candidate solutions generated is set according to the actual situation, generally 5-10. Each solution includes detailed network planning content and business decision content. For example, a candidate solution may include: network planning content, business decision content, and expected results.

[0052] Step S2.4, Scheme Evaluation and Optimization Establish a comprehensive evaluation index system to evaluate candidate solutions from multiple dimensions. The evaluation indicators include economic indicators, technical indicators, market indicators, and customer indicators. The Analytic Hierarchy Process (AHP) is used to comprehensively evaluate candidate solutions. In step S2.4, the candidate solution evaluation method is as follows: First, an evaluation team composed of network experts, business experts, market experts, and customer representatives is invited to score the importance of each evaluation indicator and determine the indicator weights. For example, for a scenario where business growth is the primary objective, the weight of the business revenue growth rate might be set to 0.3, while the weight of the network failure rate reduction rate might be set to 0.2. Then, based on actual or predicted data for each indicator, the performance of each candidate solution on each indicator is scored (out of 100). Finally, based on the indicator weights and solution scores, a comprehensive score is calculated for each candidate solution, and the solution with the highest comprehensive score is the preliminary optimal solution.

[0053] For the initial optimal solution, relevant personnel will discuss and review it to identify problems or shortcomings and propose optimization suggestions. For example, if the investment in network construction is too large, it may be necessary to adjust the construction scale or adopt a more economical technical solution; if the pricing of new services is too high, it may be necessary to appropriately reduce prices or increase package content. Based on the discussion results, the solution will be modified and improved to form the final optimal home customer network collaborative planning and decision-making solution. Step S2.5: Implementation and Monitoring of the Solution The final planning and decision-making scheme is broken down into specific implementation tasks, clearly defining the responsible departments, persons in charge, implementation timelines, and required resources for each task; and human, material, and financial resources are rationally allocated according to the implementation plan. In step S2.5, a real-time monitoring mechanism is established during the implementation of the solution. Through data acquisition equipment and business systems, the implementation progress of various tasks and the implementation effect of the solution are tracked in real time. The monitoring content includes task progress, network performance changes, business operation status and customer feedback. Real-time monitoring data is promptly fed back to the large model, which then reassesses and predicts the effectiveness of the proposed solution based on the new data. If the actual results deviate from the expected results by more than a custom threshold (e.g., a deviation exceeding 15%), the model generates suggestions for adjusting the solution. For example, if the user growth rate of a new business is significantly lower than expected, the model may suggest increasing promotional efforts or adjusting the package content.

[0054] Step S2.6, Scheme Adjustment and Iteration Based on the adjustment suggestions generated by the model and the actual monitoring situation, the implemented plan should be adjusted in a timely manner. The adjustment can be partial, such as modifying the network construction plan of a certain area or adjusting the promotional activities of a certain business; or it can be global, such as re-formulating network planning or business decision-making strategies.

[0055] After the implementation of the plan, a comprehensive summary and evaluation of the overall effect will be conducted, analyzing the successful experiences and shortcomings. Relevant data and experience will be fed back into the data warehouse and model training process to optimize the performance of the large model and improve the future collaborative planning and decision-making process of the home customer network, forming a virtuous cycle of continuous iteration.

[0056] In step S2, based on the Transformer architecture, customized improvements are made to meet the needs and goals of collaborative planning and decision-making in the home and customer network industry. A large model is constructed, and training, validation, and test sets are built. The large model is trained, validated, and optimized until it meets user needs. The specific process is as follows: Step S201: Model training data preparation (1) Dataset construction Historical data integrated from the data warehouse is used as the original dataset for model training. The dataset spans the past five years, covering data from different seasons and market environments to ensure strong generalization ability. The dataset includes network data, business data, customer data, and corresponding historical planning and decision-making data, as well as implementation results.

[0057] (2) Data partitioning The dataset is divided into training, validation, and test sets in a 7:1:2 ratio, following a chronological order. The training set contains data from the first 3.5 years, used for learning model parameters; the validation set contains data from the fourth year, used to monitor performance changes during model training and adjust hyperparameters; and the test set contains data from the last year, used to evaluate the model's final performance. Using a time-based partitioning method, rather than a random one, better reflects real-world applications where models predict the future based on historical data, allowing for more effective testing of the model's timeliness and predictive ability.

[0058] (3) Data labeling The training and validation sets are labeled with labels including network planning labels, business decision labels, and effect evaluation labels.

[0059] Data annotation employs a combination of manual and automatic annotation methods: For historical data that already contains clear planning and decision-making schemes and implementation effect records, an automatic labeling method is used to directly extract relevant information as tags; For cases where there are no clear records or where further optimization is needed, a labeling team composed of network planning experts, business analysts, and data scientists performs manual labeling, and cross-validation by multiple people ensures the accuracy of the labeling.

[0060] Step S202, Model Training Process (1) Initialize parameters The initial parameters of the large model are initialized using the parameters of a pre-trained model. The pre-trained model is a Transformer model trained on large-scale general text corpora and communication industry-related data to accelerate the model's convergence speed and improve its training performance. For custom layers in the model, such as the network data feature extraction module and the business data feature extraction module, the parameters are initialized using the Xavier initialization method.

[0061] (2) Set training parameters Learning rate: The initial learning rate is set to 5e-5, and a cosine annealing learning rate scheduling strategy is adopted. The learning rate is gradually reduced as the number of training rounds increases to avoid model oscillations in the later stages of training.

[0062] Batch size: Set to 32 based on the memory capacity of the training device, meaning that 32 data points are input simultaneously for training each time.

[0063] Training rounds: The total number of training rounds is 100. If the performance on the validation set does not improve for 10 consecutive rounds, training will be stopped early to prevent the model from overfitting.

[0064] Loss function: The cross-entropy loss function is used to calculate the loss between the model prediction result and the labeled label. For regression tasks such as effect evaluation, the mean squared error loss function is used. The total loss is the weighted sum of the classification loss and the regression loss. The weights are customized according to the importance of the task. In this invention, the classification loss weight is 0.6 and the regression loss weight is 0.4.

[0065] (3) Model training and monitoring A distributed training framework is used, leveraging 8 GPU servers for parallel training to improve training efficiency. During training, after each training round, the large model is evaluated using a validation set, calculating metrics such as prediction accuracy and mean squared error, and recording model parameters. Simultaneously, the model's loss on the training and validation sets is monitored. If the loss on the training set continues to decrease while the loss on the validation set increases, it indicates that the large model is overfitting. In this case, an early stopping strategy is employed to halt training, and the parameters of the model with the best performance on the validation set are loaded.

[0066] Step S203, Model Evaluation and Optimization (1) Evaluation indicators The following metrics were used to evaluate model performance: accuracy, root mean square error (RMSE), and decision performance metrics.

[0067] (2) Model optimization Based on the evaluation results, the large model was optimized in a targeted manner: If the model performs poorly on a certain type of data, such as the accuracy of network planning prediction in rural areas being lower than a custom threshold, then increase the proportion of that type of data in the training set and perform targeted fine-tuning training. If the large model exhibits overfitting, dropout regularization is employed by adding dropout layers to the encoding and decoding layers with a dropout rate of 0.1 to reduce excessive dependence between neurons. If the convergence speed of a large model is slow, adjust the learning rate scheduling strategy or increase the batch size to speed up the training process of the model. If the prediction accuracy of the large model does not meet the user's requirements, the depth and width of the large model can be increased, such as by increasing the number of encoder and decoder layers and expanding the number of attention heads, in order to improve the model's feature extraction and representation capabilities.

[0068] After multiple rounds of evaluation and optimization, the final large model achieved an accuracy of over 90% on the test set, with the root mean square error (RMSE) controlled within 5% and the deviation of the decision performance index less than 10%, meeting the requirements of practical applications. Training was then stopped, and the trained large model was obtained.

[0069] In step S2, the continuous optimization and maintenance process of the large model is as follows: (1) Model performance monitoring Establish a model performance monitoring system to regularly (e.g., daily) evaluate the performance of large models in real-world applications. Monitoring metrics include prediction accuracy, decision performance bias, and response time. If a significant decline in model performance is observed (e.g., accuracy drops by more than 10%), analyze the causes promptly.

[0070] (2) Data update and retraining As time goes on, customer networks and business data accumulate, and market conditions and user needs also change. Therefore, it is necessary to update the training data regularly (e.g., quarterly), adding new historical data and implementation effect data to the training set to retrain the large model and maintain its timeliness and accuracy.

[0071] (3) Model version management The system manages different versions of a large model, recording the training data, model parameters, performance metrics, and application scenarios for each version. When a model is updated or optimized, a new version is generated and compared with the old version to ensure that the new version outperforms the old version before being deployed in real-world applications. Simultaneously, the model parameters and data from older versions are retained for backtracking and analysis when needed.

[0072] (4) Security and privacy protection Throughout the training, inference, and application of large models, we strictly adhere to relevant laws and regulations regarding data security and privacy protection. We employ encryption technologies to protect sensitive data and prevent data leakage. Access to the models is managed with access permissions, ensuring only authorized personnel can access and use them. Regular security vulnerability checks and risk assessments are conducted to ensure the secure and stable operation of the models.

[0073] This large-scale model-based collaborative planning and decision-making system for the home and customer network industry is used to implement the above methods, including: Data collection module: Used to collect network data, business data, and customer data, including network data collection submodule, business data collection submodule, and customer data collection submodule; Among them, the network data acquisition submodule is responsible for collecting topology and performance data through the network management system and monitoring equipment, the business data acquisition submodule is responsible for obtaining business data from the business support system and complaint management system, and the customer data acquisition submodule is responsible for integrating customer information in the customer relationship management system. Data processing module: Used to process the collected data, including data cleaning unit, standardization unit, missing value imputation unit and data integration unit; Among them, the data cleaning unit is responsible for removing duplicate and erroneous data, the standardization unit is responsible for normalizing and encoding the data, the missing value filling unit is responsible for filling missing data using multiple methods, and the data integration unit is responsible for integrating the processed data into the data warehouse. Large Model Training Module: Used to build, train, and optimize large models, including model architecture design unit, training data preparation unit, parameter setting unit, model training unit, and model evaluation and optimization unit; The model architecture design unit is responsible for designing a large model structure based on Transformer; the training data preparation unit is responsible for building and partitioning the dataset and labeling it; the parameter setting unit is responsible for setting the training parameters; the model training unit is responsible for training the model using the training set; and the model evaluation and optimization unit is responsible for evaluating the model performance and optimizing it.

[0074] Collaborative Planning and Decision-Making Module: Used to execute the collaborative planning and decision-making process of the home customer network industry, including a demand target setting unit, a data input preprocessing unit, a solution generation unit, a solution evaluation and optimization unit, a solution implementation monitoring unit, and a solution adjustment and iteration unit; The unit is responsible for collecting requirements and setting targets; the data input preprocessing unit is responsible for inputting and preprocessing real-time data; the solution generation unit is responsible for generating candidate solutions using a large model; the solution evaluation and optimization unit is responsible for evaluating and optimizing solutions; the solution implementation monitoring unit is responsible for developing plans and monitoring implementation; and the solution adjustment and iteration unit is responsible for adjusting solutions and providing data feedback.

[0075] The large-scale model, based on the Transformer architecture, comprises an input layer, an encoding layer, a fusion layer, a decoding layer, and an output layer. The input layer receives network data, business data, and customer data, respectively, and processes them through their respective feature extraction modules. The encoding layer consists of multiple Transformer encoders that perform deep encoding on the extracted features. The fusion layer uses a cross-attention mechanism to fuse different types of features. The decoding layer consists of multiple Transformer decoders that generate the planning and decision-making scheme. The output layer outputs the final planning and decision-making scheme and evaluation metrics.

[0076] The large-scale model-based collaborative planning and decision-making device for home and customer networks includes a memory and a processor; the memory is used to store computer programs, and the processor is used to execute the computer programs to implement the above-described method steps.

[0077] The readable storage medium stores a computer program that, when executed by a processor, implements the above-described method steps.

[0078] Compared with existing technologies, this large-model-based collaborative planning and decision-making method for home and customer networks has the following characteristics: First, it significantly improved economic efficiency: (1) Reduced operating costs Through precise planning using large-scale models, excessive investment and waste of network resources are avoided.

[0079] (2) Increase business revenue The method of this invention can formulate precise business decision-making strategies based on user needs and network conditions, effectively improving business revenue.

[0080] (3) Improved return on investment The collaborative planning decision-making schemes generated by the large model have undergone comprehensive evaluation and optimization, ensuring a high return on investment.

[0081] Secondly, network performance and service quality have been optimized: (1) Improved network stability By using large models to analyze and predict network performance data in real time, potential faults and bottlenecks in the network can be identified in a timely manner, and measures can be taken in advance to deal with them.

[0082] (2) Improved network service quality The large model optimizes network resource allocation based on user business needs and network conditions, resulting in a significant improvement in network service quality.

[0083] (3) Improve the utilization rate of theoretical network resources Through precise network planning, network resources are rationally allocated, and resource utilization is significantly improved.

[0084] Third, it enhanced customer experience and market competitiveness: (1) Improved customer satisfaction This invention provides customers with higher-quality and more personalized services, effectively improving customer satisfaction. Through in-depth analysis of customer needs using a large-scale model, operators can launch service packages and content that meet customer requirements.

[0085] (2) Enhanced market competitiveness By responding quickly to market changes and user needs, and launching competitive services in a timely manner, operators have significantly enhanced their market competitiveness.

[0086] (3) Promoted business innovation Large models can uncover potential user needs and market opportunities, providing inspiration and direction for business innovation.

[0087] Fourth, it improved decision-making efficiency and management level: (1) Shortened the decision-making cycle Traditional business planning and decision-making for home appliances requires lengthy discussions and analyses across multiple departments, typically taking 1-2 months. By adopting the method of this invention, through automated analysis and solution generation using large-scale models, the decision-making cycle is shortened to 1-2 weeks, significantly improving the timeliness and efficiency of decision-making. For example, in responding to sudden network failures or market changes, effective solutions can be quickly developed to minimize losses.

[0088] (2) Improved the scientific nature and accuracy of decision-making. The large-scale model analyzes and predicts based on massive amounts of data and advanced algorithms, avoiding the subjectivity and blindness of human decision-making and making decisions more scientific and accurate. For example, when predicting the market prospects of a new business, the prediction error of traditional methods may be as high as 30%, while the prediction error of the method of this invention can be controlled within 10%, providing a reliable basis for business decisions.

[0089] (3) Promoted inter-departmental collaboration This invention integrates the needs and data of multiple departments, enabling information sharing and collaborative work among them. During the planning and decision-making process, departments can communicate and discuss based on unified data and models, reducing inter-departmental conflicts and improving the overall management level and work efficiency of the enterprise.

[0090] The embodiments described above are merely one specific implementation of the present invention. Ordinary changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included within the protection scope of the present invention.

Claims

1. A collaborative planning and decision-making method for home and customer networks based on a large model, characterized in that: Includes the following steps: Step S1: Data Collection and Processing Collect network data, business data, and customer data from the home customer network, and clean the collected data to remove duplicate, erroneous, and incomplete data records; Data standardization methods are used to normalize data of different magnitudes and units; Fill in missing data; integrate the processed data into a unified data warehouse and establish relationships; Step S2: Home Customer Network Industry Collaborative Planning and Decision-Making Process Customize the planning and decision-making needs and objectives, and set quantitative objectives that conform to the SMART principle; preprocess the current real-time network data, real-time business data, and real-time customer data, and then input them into the trained large model; The large model performs inference calculations on the input data to generate several candidate home-based business collaborative planning and decision-making schemes; The analytic hierarchy process (AHP) is used to comprehensively evaluate candidate solutions based on evaluation indicators to determine the optimal solution. Break down the optimal solution into specific implementation tasks, formulate an implementation plan, and allocate resources for implementation; During implementation, task progress, network performance changes, business operations, and customer feedback are monitored in real time, and the feedback data is input into the large model. Based on the model evaluation results and actual monitoring, the solution is adjusted and iterated, and the relevant data is fed back into the data warehouse and the model training process.

2. The home-customer network collaborative planning and decision-making method based on a large model as described in claim 1, characterized in that: In step S1, the network data of the home customer network includes topology, performance data and equipment configuration data; the service data includes the usage of home customer services, complaint data and competition data; and the customer data includes basic customer information and internet access behavior data.

3. The home-customer network collaborative planning and decision-making method based on a large model as described in claim 1, characterized in that: In step S2, the collaborative planning and decision-making process of Jiake.com is as follows: Step S2.1: Requirements Analysis and Goal Setting Before making collaborative planning decisions, collect the needs and expectations of each department and set clear and quantifiable planning and decision-making goals; Step S2.2, Data Input and Preprocessing Preprocess the real-time data input to the large model, including data cleaning, standardization, and integration, to ensure that the quality and format of the input data are consistent with the requirements of the large model; The current real-time data is input into the trained large model, including real-time network data, real-time business data, and real-time customer data; Step S2.3, Model Reasoning and Solution Generation The preprocessed real-time data and the set planning and decision-making objectives are input into the trained large model, which then performs inference and calculation to generate several candidate home-customer network collaborative planning and decision-making schemes. During the inference process, the model comprehensively considers network status, business needs, customer preferences and market competition factors in real-time data, as well as the set quantitative targets, to conduct multi-dimensional analysis and prediction. The number of candidate solutions generated is 5-10, and each solution contains detailed network planning and business decision-making content; Step S2.4, Scheme Evaluation and Optimization Establish a comprehensive evaluation index system to evaluate candidate solutions from multiple dimensions. The evaluation indicators include economic indicators, technical indicators, market indicators, and customer indicators. The Analytic Hierarchy Process (AHP) is used to comprehensively evaluate candidate solutions. Step S2.5: Implementation and Monitoring of the Solution The final planning and decision-making scheme is broken down into specific implementation tasks, clearly defining the responsible departments, persons in charge, implementation timelines, and required resources for each task; and human, material, and financial resources are allocated according to the implementation plan. Step S2.6, Scheme Adjustment and Iteration Based on the adjustment suggestions generated by the model and the actual monitoring situation, the implemented plan is adjusted in a timely manner; After the implementation of the plan, a comprehensive summary and evaluation of the overall effect will be conducted, analyzing the successful experiences and shortcomings. Relevant data and experience will be fed back into the data warehouse and model training process to optimize the performance of the large model and improve the future collaborative planning and decision-making process of the home customer network, forming a virtuous cycle of continuous iteration.

4. The home-customer network collaborative planning and decision-making method based on a large model as described in claim 3, characterized in that: In step S2.4, the candidate solution evaluation method is as follows: First, an evaluation team composed of network experts, business experts, market experts, and customer representatives is invited to score the importance of each evaluation indicator and determine the indicator weights. Then, based on the actual or predicted data of each indicator, the performance of each candidate solution on each indicator is scored. Finally, based on the indicator weights and solution scores, the comprehensive score of each candidate solution is calculated, and the solution with the highest comprehensive score is the preliminary optimal solution. For the preliminary optimal solution, relevant personnel are organized to discuss and review it, identify problems or deficiencies in the solution, and propose optimization suggestions; based on the discussion results, the solution is modified and improved to form the final optimal home-customer network collaborative planning and decision-making solution.

5. The home-customer network collaborative planning and decision-making method based on a large model as described in claim 3, characterized in that: In step S2.5, a real-time monitoring mechanism is established during the implementation of the solution. Through data acquisition equipment and business systems, the implementation progress of various tasks and the implementation effect of the solution are tracked in real time. The monitoring content includes task progress, network performance changes, business operation status and customer feedback. Real-time monitoring data is fed back to the large model in a timely manner. The model re-evaluates and predicts the implementation effect of the plan based on the new data. If the deviation between the actual effect and the expected effect exceeds a custom threshold, the model generates suggestions for adjusting the plan.

6. The home-customer network collaborative planning and decision-making method based on a large model according to claim 1, characterized in that: In step S2, based on the Transformer architecture, customized improvements are made to meet the needs and objectives of collaborative planning and decision-making in the home and customer network industry, and a large model is constructed. The integrated historical data in the data warehouse is used as the original dataset for model training. The dataset is divided into training set, validation set and test set in chronological order, with a ratio of 7:1:

2. The training and validation sets are labeled with labels including: network planning labels, business decision labels, and effect evaluation labels. Data annotation employs a combination of manual and automatic annotation methods: For historical data that already contains clear planning and decision-making schemes and implementation effect records, an automatic labeling method is used to directly extract relevant information as tags; For cases where there are no clear records or where further optimization is needed, a labeling team composed of network planning experts, business analysts, and data scientists performs manual labeling, and cross-validation by multiple people ensures the accuracy of the labeling. The large model is trained, validated, and optimized; the model training process is as follows: (1) Initialize parameters The initial parameters of the large model are initialized using the parameters of the pre-trained model. The pre-trained model is a Transformer model trained on large-scale general text corpora and communication industry-related data. For the custom layers in the model, the parameters are initialized using the Xavier initialization method. (2) Set training parameters Learning rate: The initial learning rate is set to 5e-5, and a cosine annealing learning rate scheduling strategy is adopted. The learning rate is gradually reduced as the number of training rounds increases. Batch size: Set to 32 based on the memory capacity of the training device, meaning that 32 data points are input for training at the same time each time; Training rounds: The total number of training rounds is 100. If the performance on the validation set does not improve for 10 consecutive rounds, training will be stopped early to prevent the model from overfitting. Loss function: The cross-entropy loss function is used to calculate the loss between the model prediction result and the labeled label. For regression tasks, the mean squared error loss function is used. The total loss is the weighted sum of the classification loss and the regression loss, with a classification loss weight of 0.6 and a regression loss weight of 0.

4. (3) Model training and monitoring Parallel training is performed using a distributed training framework. During training, after each round of training, the large model is evaluated using a validation set. The prediction accuracy, mean squared error, and other metrics of the large model are calculated, and the model parameters are recorded. At the same time, the changes in the model's loss on the training and validation sets are monitored. If the loss on the training set continues to decrease while the loss on the validation set increases, it indicates that the large model is overfitting. In this case, an early stopping strategy is adopted to stop training, and the model parameters with the best performance on the validation set are loaded. The model evaluation and optimization process is as follows: (1) Set evaluation indicators The following metrics were used to evaluate the model performance: accuracy, root mean square error, and decision performance. (2) Model optimization Based on the evaluation results, the large model was optimized in a targeted manner: If the model's evaluation metric on a certain type of data is lower than a custom threshold, then increase the proportion of that type of data in the training set and perform targeted fine-tuning training. If the large model exhibits overfitting, dropout regularization is employed by adding dropout layers to the encoding and decoding layers with a dropout rate of 0.1 to reduce excessive dependence between neurons. If the convergence speed of a large model is slow, adjust the learning rate scheduling strategy or increase the batch size to speed up the training process of the model. If the prediction accuracy of the large model does not meet the user's requirements, the depth and width of the large model are increased to improve the model's feature extraction and representation capabilities. After multiple rounds of evaluation and optimization, the final large model achieved an accuracy of over 90% on the test set, with the root mean square error controlled within 5%. Training was stopped when the deviation of the decision performance index was less than 10%, resulting in the trained large model.

7. A collaborative planning and decision-making system for home and customer networks based on a large model, characterized in that: To implement the method according to any one of claims 1 to 6, comprising: Data collection module: Used to collect network data, business data, and customer data, including network data collection submodule, business data collection submodule, and customer data collection submodule; Among them, the network data acquisition submodule is responsible for collecting topology and performance data through the network management system and monitoring equipment, the business data acquisition submodule is responsible for obtaining business data from the business support system and complaint management system, and the customer data acquisition submodule is responsible for integrating customer information in the customer relationship management system. Data processing module: Used to process the collected data, including data cleaning unit, standardization unit, missing value imputation unit and data integration unit; Among them, the data cleaning unit is responsible for removing duplicate and erroneous data, the standardization unit is responsible for normalizing and encoding the data, the missing value filling unit is responsible for filling missing data using multiple methods, and the data integration unit is responsible for integrating the processed data into the data warehouse. Large Model Training Module: Used to build, train, and optimize large models, including model architecture design unit, training data preparation unit, parameter setting unit, model training unit, and model evaluation and optimization unit; Among them, the model architecture design unit is responsible for designing the large model structure based on Transformer, the training data preparation unit is responsible for building and dividing the dataset and labeling it, the parameter setting unit is responsible for setting the training parameters, the model training unit is responsible for training the model using the training set, and the model evaluation and optimization unit is responsible for evaluating the model performance and optimizing it. Collaborative Planning and Decision-Making Module: Used to execute the collaborative planning and decision-making process of the home customer network industry, including a demand target setting unit, a data input preprocessing unit, a solution generation unit, a solution evaluation and optimization unit, a solution implementation monitoring unit, and a solution adjustment and iteration unit; The unit is responsible for collecting requirements and setting targets; the data input preprocessing unit is responsible for inputting and preprocessing real-time data; the solution generation unit is responsible for generating candidate solutions using a large model; the solution evaluation and optimization unit is responsible for evaluating and optimizing solutions; the solution implementation monitoring unit is responsible for developing plans and monitoring implementation; and the solution adjustment and iteration unit is responsible for adjusting solutions and providing data feedback.

8. A collaborative planning and decision-making device for home and customer networks based on a large model, characterized in that: It includes a memory and a processor; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the method according to any one of claims 1 to 6.

9. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 6.

Images