Heat supply secondary network personalized regulation and control method based on large model and multi-modal portrait

Abstract

The invention discloses a heat supply secondary network personalized regulation and control method based on a large model and a multi-modal portrait, and the method comprises the steps: obtaining the multi-modal data of each heat user of a heat supply system secondary network, carrying out the feature extraction and importance analysis through a large model, and constructing each heat user portrait of the secondary network; performing information extraction on each household operation mode of the secondary network of the heat supply system, combining and selecting a heat user portrait of the household operation mode, and generating an operation mode portrait of the household operation mode by utilizing a large model; inputting the portraits of the heat users and the operation mode portraits of the secondary network into the large model for matching analysis, and outputting personalized operation mode recommendation results and recommendation verbal skills for the heat users; and after an operation mode is recommended and selected to the heat user, according to the operation mode selected by the heat user, the heat user portrait and the established household regulation and control knowledge graph, in combination with heat supply network hydraulic balance mechanism knowledge, a large heat supply system secondary network household regulation and control model is trained and optimized, and a secondary network household regulation and control strategy is output.

Description

Technical Field

[0001] The present invention belongs to the technical field of secondary network regulation of heating systems, and specifically relates to a personalized regulation method for heating secondary networks based on large models and multimodal portraits. Background Art

[0002] The main task of the secondary network of the heating system is to distribute the fluid in the primary pipe network to specific areas or heat users. The operation of the secondary pipe network is directly related to the heat demand of heat-using residents, enterprises, and commercial areas. Different types of heat users have different heat-using behaviors and preference attributes. For example, some residential heat users are not at home during the day and use heat after work. Then, this resident can choose to lower the indoor temperature and reduce the heat consumption during the period when no one is at home during the day; for commercial heat users such as shopping malls, their heat-using periods are related to business hours, and the heat consumption is related to the population density. Therefore, it is necessary to establish portraits of each heat user in the secondary network to characterize the heat-using behaviors and preferences of different heat users, so as to facilitate personalized regulation.

[0003] With the application of artificial intelligence technology in the field of intelligent heating, the optimal regulation of the heating system has become more intelligent. However, with the growth of heat user data volume in the heating system, the expansion of the system operation scale, and the complex diversity of operation data, etc., how to efficiently and accurately process and analyze these data, generate better regulation strategies, and meet the heat-using needs of different heat users has become a difficult problem. Although machine learning and deep learning technologies have been applied to heating optimal regulation, these methods still have certain limitations, lack personalization, are difficult to reveal the complex correlations between multimodal massive heterogeneous data, cannot output personalized regulation strategies for each heat user, and have not fully utilized the rich structured knowledge contained in the regulation process of the heating secondary network.

[0004] Based on the above technical problems, it is necessary to design a new personalized regulation method for heating secondary networks based on large models and multimodal portraits. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a personalized regulation method for heating secondary networks based on large models and multimodal portraits, which can use large models and multimodal data to establish portraits of each heat user and operation mode portraits in the secondary network, conduct matching analysis, output personalized operation mode recommendation results and recommended words for heat users, clearly explain the recommendation reasons and mode advantages to users, improve the heat user experience and acceptance of the operation mode; at the same time, establish a household regulation knowledge graph, make full use of the structured knowledge of regulation, establish a household regulation large model, mine the complex correlations between data, make the model continuously adapt to the changes in the actual situation, and improve the accuracy and effectiveness of the regulation strategy.

[0006] To solve the above technical problems, the technical solution of the present invention is:

[0007] The present invention provides a personalized regulation method for the second-level heating network based on a large model and multimodal portraits, which includes:

[0008] S1. Obtain the multimodal data of each heat user in the second-level heating network, and after performing feature extraction and importance analysis using a large model, construct portraits of each heat user in the second-level heating network;

[0009] S2. Extract information on the operation modes of each household in the second-level heating network, combine with the portraits of the heat users who select the operation modes of each household, and use a large model to generate operation mode portraits for the operation modes of each household;

[0010] S3. Input the portraits of each heat user and the operation mode portraits in the second-level heating network into the large model for matching analysis, and output personalized operation mode recommendation results and recommendation scripts for each heat user;

[0011] S4. Establish a household regulation knowledge graph based on the household heat metering data, regulation data of each heat user, and regulation knowledge corresponding to the operation modes;

[0012] S5. According to the operation mode recommendation results and recommendation scripts of each heat user, recommend to the heat users to select an operation mode, and then, according to the operation mode selected by the heat user, the heat user portrait, and the household regulation knowledge graph, combined with the knowledge of the heat network hydraulic balance mechanism, train and optimize the household regulation large model of the second-level heating network, and output the household regulation strategy for the second-level heating network.

[0013] Further, the specific steps of S1 include:

[0014] Obtain the multimodal data of each heat user in the second-level heating network, including text data, voice data, and image data; the text data includes basic information such as heat user type, user age, house orientation, house area, floor where the user is located, as well as user's work and rest time, heat consumption in each period, flexible heat load response information, complaint information, maintenance work order information, and preferred operation mode; the voice data includes voice evaluation information on heating effect, fault description voice information, and special heat use demand voice information between the heat user and the heating enterprise; the image data includes images of the indoor heating equipment status, outdoor heating pipeline images, house exterior images, house layout images, and images of people flowing in the house; the heat user type includes civil heating and industrial and commercial heating;

[0015] Preprocess and perform multimodal fusion on the acquired multimodal data to form a structured dataset, and input it into a pre-trained large model. Using the method of combining the large model and prompt words, take the structured dataset and prompt words as input, and then use the attention mechanism to extract features from the data, including feature extraction of text, speech, and images, to obtain the basic features of hot users and the hot behavior features of users. After performing feature importance analysis using the feature importance evaluation method, output the hot user portraits of each secondary network including multiple portrait dimensions.

[0016] Furthermore, the prompt words when constructing the hot user portraits of each secondary network include:

[0017] Role setting: Clearly define that the large model is an expert in the heating field during this conversation process, guiding the large model to output the behavior patterns and response methods of hot user portraits;

[0018] Requirement description: Describe the task requirements completed by the large model, including feature extraction and feature importance analysis;

[0019] Definition of multiple portrait dimensions: Include the definition of basic user features and the definition of user hot behavior, guiding the large model to correctly understand specific meanings; among them, the definition of basic user features is the static information of hot users; the definition of user hot behavior is the dynamic information of hot users, including hot user daily routine behavior information, hot user heat consumption preference information for each time period, flexible response information, operation mode preference information, and heat consumption feedback information.

[0020] Furthermore, the hot user portraits of each secondary network are defined as:

[0021] P i = LLMs(D i , R);

[0022] D i is the structured dataset of the i-th hot user; R is the prompt word when constructing the hot user portrait; P i is the portrait of the i-th hot user; LLMs() is the large model used when constructing the hot user portrait.

[0023] Furthermore, in S2, the specific operation modes of each household in the secondary network of the heating system include:

[0024] Conventional operation mode: At any time during the heating season, the indoor temperature of the user reaches the preset temperature;

[0025] Energy-saving operation mode: It has a time period division mechanism and a customized temperature selection mechanism, and can select to save energy by setting a lower indoor temperature than the conventional preset temperature during a fixed time period according to the user's heat consumption behavior, and at the same time give economic compensation;

[0026] Intelligent Adaptive Mode: It can monitor the indoor temperature, room activity time, and the number of flowing people in real time, and adjust the heating supply intelligently and automatically.

[0027] Economic Optimization Mode: According to the energy price fluctuation information, on the premise of meeting the basic comfort level, it provides the optimal operation plan to reduce the heating cost.

[0028] Furthermore, the specific steps of S2 include:

[0029] Obtain the definitions of the operation modes of each household in the secondary network of the heating system, the heat user information of each household's historical selected operation modes, and the operation parameters of the operation modes.

[0030] Set the prompt words for guiding the large model to generate the operation mode portrait, and clarify its function in the description of the operation mode portrait generation.

[0031] Take the name of the operation mode, the description of the mode characteristics, the association information between each operation mode and the heat user portrait, and the operation parameters of the operation mode as the operation mode data, and input them into the large model together with the prompt words to generate the operation mode portrait, expressed as:

[0032] P j = LLMs′(L j , M);

[0033] L j is the operation mode data of the j-th operation mode; M is the prompt word when generating the operation mode portrait; P j is the portrait of the j-th operation mode; LLMs′ is the large model used when generating the operation mode portrait.

[0034] Furthermore, the large model generating the operation mode portrait includes:

[0035] Use the tokenizer to split the input sequence data into tokens, and map each token to a low-dimensional vector space to obtain token embeddings. At the same time, add position encoding to let the large model learn the position information of the tokens in the sequence.

[0036] Based on the multi-layer transformer encoder structure of the large model, use the multi-head self-attention mechanism of each encoder to enable the large model to learn the relationship between tokens in different representation subspaces, capture semantic information, and then use the feed-forward neural network to perform information transformation and feature extraction on the output of the multi-head self-attention, learning different abstract representations of the input data at different levels. Through continuous stacking, obtain the feature representation containing rich information of the input data.

[0037] At the output layer, a linear transformation is used to map this feature representation to a dimension with the same size as the vocabulary size to obtain the predicted scores for each token; a decoding strategy is used to select the token with the highest probability according to the predicted scores as the next generated token, and gradually generate the text content of the operation mode portrait.

[0038] Further, the specific steps of S3 include:

[0039] Integrate the portrait of each hot user and the operation mode portrait in the secondary network into an input data set;

[0040] Set prompts to guide the large model to output personalized operation mode recommendation results and recommendation scripts, and clarify the role of the large model, task requirement descriptions, and thinking chain construction during operation mode recommendation and script recommendation;

[0041] Input the input data set and prompts into the large model, and the large model uses internal attention mechanisms and neural network layers to calculate the similarity between the hot user portrait features and the features of each operation mode portrait;

[0042] Based on the similarity calculation results, the large model selects the operation mode with the highest similarity as the personalized recommendation result for hot users to output;

[0043] Based on the output personalized operation mode recommendation results, the large model combines the pre-established script corpus to output differentiated recommendation scripts for different hot users.

[0044] Further, the specific steps of S4 include:

[0045] Obtain the heat metering data of each hot user, including heat consumption, flow rate, and temperature in different time periods;

[0046] Obtain the household control data of each hot user, including the operation mode switching time and the adjustment parameters of the control equipment;

[0047] Obtain the control knowledge corresponding to the operation mode, including: the control mechanism of the control equipment and the target temperature range under different operation modes;

[0048] Based on the household heat metering data, control data, and control knowledge corresponding to the operation mode of each hot user, use deep learning methods to perform entity, relationship, and attribute extraction respectively; the relationship extraction includes the relationship between the hot user and the control equipment, the relationship between the hot user and the operation mode, and the relationship between the control equipment and the operation mode;

[0049] Build a household regulation knowledge graph based on the extracted entities, relationships, and attributes, including: determining the structure of the household regulation knowledge graph for the secondary network, in the form of a directed graph, with nodes representing entities and edges representing the relationships between entities; adding entities as nodes to the knowledge graph, creating corresponding edges according to the relationships, and adding corresponding attribute information to each node and edge.

[0050] Further, the specific steps of S5 include:

[0051] Associate and integrate the operation mode data selected by the heat users with the heat user portrait data;

[0052] Extract knowledge information related to the heat user operation mode and regulation equipment from the household regulation knowledge graph, convert it into a vector form suitable for model input, and fuse the processed knowledge graph vector with the heat user data;

[0053] Convert the heat network hydraulic balance mechanism knowledge into a mathematical model and constraint conditions, and extract the parameters related to the heat network hydraulic balance;

[0054] Define the training objective: aiming to generate the optimal household regulation strategy for the secondary network, set the training objective function according to the operation mode requirements of the heat users, minimize the heat network energy consumption, and ensure the heat network hydraulic balance;

[0055] Integrate the associated data of the operation mode and heat user portrait, the knowledge graph data, and the heat network hydraulic balance data as training samples and input them into the large model for training to generate a large model for household regulation of the secondary network of the heating system, and continuously adjust the parameters of the large model so that the large model can learn the relationships among the heat user operation mode, heat user portrait, household regulation knowledge graph, and heat network hydraulic balance;

[0056] For the large model after training and optimization, input the new heat user operation mode, heat user portrait, and real-time operation data of the secondary network. The large model generates the household regulation strategy for the secondary network according to the learned knowledge and relationships, including the adjustment parameters of the pre-household regulation equipment and the switching time of the operation mode.

[0057] The beneficial effects of the present invention are:

[0058] (1) By obtaining multi-modal data and using the large model for feature extraction and importance analysis to construct the heat user portrait, the present invention can comprehensively depict the characteristics and heat consumption characteristics of heat users using multi-modal data, and can accurately extract key features from a large amount of multi-modal data by using the powerful feature extraction ability of the large model, filter out redundant information, and at the same time determine the influence degree of each feature on the heat consumption behavior of users through importance analysis. The constructed heat user portrait provides the basic user information for subsequent personalized operation mode recommendation and regulation;

[0059] (2) The present invention extracts household operation mode information, combines it with thermal user portraits, and uses a large model to generate operation mode portraits, which can sort out the key information and characteristics of each household operation mode in detail, and combine the operation mode with the thermal user portrait that selects the mode, so that the operation mode portrait is more in line with actual user needs and applicable scenarios, which facilitates understanding of the user group characteristics applicable to each mode, and helps to more accurately recommend appropriate operation modes for thermal users;

[0060] (3) The present invention inputs the heat user portrait and the operation mode portrait into the big model for matching analysis, and outputs personalized recommendation results and words. It can use the powerful data analysis and matching capabilities of the big model to accurately find the operation mode that best suits each heat user based on the heat user portrait and the operation mode portrait, thereby improving the accuracy of the recommendation. In addition, according to the characteristics and needs of different heat users, the present invention outputs personalized recommendation words, clearly explains the reasons for the recommendation and the advantages of the mode to the user, and improves the heat user experience and acceptance of the operation mode.

[0061] (4) The present invention establishes a household regulation knowledge graph based on household heat metering data, regulation data and regulation knowledge, and can integrate various types of data and regulation knowledge to form a structured knowledge graph, which is convenient for unified management and query of household regulation related information; in addition, it provides rich knowledge support for household regulation of the secondary network of the heating system, which helps the large model to comprehensively consider various factors when generating regulation strategies and make more scientific and reasonable decisions; and through the association analysis of the knowledge graph, it can discover the potential relationship between household heat metering data, regulation data and regulation knowledge, providing a basis for optimizing the regulation strategy;

[0062] (5) The present invention guides users to select a mode based on recommendation results and words, combines various information to train and optimize the large control model, and outputs a control strategy. It can train and optimize the large control model based on various information such as the operation mode selected by the user, the heat user portrait, and the household control knowledge map, combined with the knowledge of the hydraulic balance mechanism of the heating network, so that the model can continuously adapt to changes in actual conditions and improve the accuracy and effectiveness of the control strategy. In addition, the knowledge of the hydraulic balance mechanism of the heating network is considered to ensure that the output secondary network household control strategy can maintain the hydraulic balance of the heating network while meeting the personalized needs of users, thereby ensuring the stable operation of the entire heating system.

[0063] Other features and advantages will be described in the following description, and partly become apparent from the description, or understood by practicing the invention. The purpose and other advantages of the invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0064] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. Description of the Drawings

[0065] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0066] Figure 1 It is a flowchart of a personalized regulation method for the second-level heating network based on a large model and multi-modal portraits of the present invention;

[0067] Figure 2 It is a principle block diagram of the personalized regulation of the second-level heating network based on a large model and multi-modal portraits of the present invention;

[0068] Figure 3 It is a flowchart of a method for outputting personalized operation mode recommendation results and recommendation scripts of the present invention;

[0069] Figure 4 It is a flowchart of a method for establishing a household regulation knowledge graph of the present invention;

[0070] Figure 5 It is a flowchart of a method for training and optimizing a large model for household regulation of the second-level heating network of the present invention. Detailed Embodiments

[0071] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions of the present invention with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention fall within the scope of protection of the present invention.

[0072] As Figure 1 and Figure 2 shown, Embodiment 1 of the present invention provides a personalized regulation method for the second-level heating network based on a large model and multi-modal portraits, which includes:

[0073] S1. Obtain multi-modal data of each heat user in the second-level heating network of the heating system, and construct portraits of each heat user in the second-level heating network after feature extraction and importance analysis using a large model;

[0074] S2. Extract information on the operation modes of each household in the second-level heating network of the heating system, combine it with the portraits of the heat users who select the operation modes of each household, and generate operation mode portraits of the operation modes of each household using a large model;

[0075] S3. Input the portraits of each heat user and the operation mode portrait in the secondary network into the large model for matching analysis, and output personalized operation mode recommendation results and recommendation scripts for each heat user;

[0076] S4. Based on the household heat metering data, regulation data of each heat user, and regulation knowledge corresponding to the operation mode, establish a household regulation knowledge graph;

[0077] S5. Based on the operation mode recommendation results and recommendation scripts of each heat user, recommend to the heat user to select an operation mode. Then, according to the operation mode selected by the heat user, the heat user portrait, and the household regulation knowledge graph, combined with the knowledge of the hydraulic balance mechanism of the heat network, train and optimize the household regulation large model of the secondary network of the heating system, and output the household regulation strategy of the secondary network.

[0078] In this embodiment, the S1 specifically includes:

[0079] Obtain multi-modal data of each heat user in the secondary network of the heating system, including text data, voice data, and image data; the text data includes basic information such as heating user type, user age, house orientation, house area, and floor where the user is located, as well as user's work and rest time, heat consumption in each period, flexible heat load response information, complaint information, maintenance work order information, and preferred operation mode; the voice data includes voice evaluation information on heating effect, voice information on fault description, and voice information on special heat use requirements between the heat user and the heating enterprise; the image data includes images of the indoor heating equipment status, outdoor heating pipeline images, house exterior images, house layout images, and images of people flowing in the house; the heating user type includes civil heating and industrial and commercial heating;

[0080] Preprocess and perform multi-modal fusion on the obtained multi-modal data to form a structured data set, and input it into the pre-trained large model. Use the method of combining the large model and prompt words, take the structured data set and prompt words as inputs, and then use the attention mechanism to extract features from the data, including feature extraction of text, voice, and image, to obtain the basic features of the heat user and the heat use behavior features of the user. After performing feature importance analysis using the feature importance evaluation method, output the portraits of each heat user in the secondary network including multiple portrait dimensions.

[0081] In actual applications, for civil heating, different residential users have different preferences for heating behaviors. For example, young people have irregular schedules, are active at night, and are not at home during the day. They can try the energy-saving heating mode. When there is no one at home during the day, the indoor temperature can be lowered to flexibly adjust the heat load accordingly, and at the same time, some economic compensation can be obtained. When they get off work and return home, the heating operation mode is switched to ensure the indoor heating demand. The elderly and children have relatively regular home times, relatively weak constitutions, low tolerance to cold, and high requirements for room temperature. They pay attention to comfort and generally do not frequently adjust the temperature. For different types of industrial and commercial users, the heating behavior preferences are also different. For example, for production-type industries, different production processes have specific requirements for heat environment parameters such as temperature and humidity. The heating demand is usually relatively stable and large, and heating interruption may cause serious losses such as production stagnation and product scrapping. For food processing enterprises, heating has periodicity, with increased heating during the peak production season and decreased heating during the off-season. For shopping malls and shopping centers, the business hours are fixed and long. During business hours, a comfortable indoor temperature needs to be maintained to attract customers and improve the shopping experience. The passenger flow is large on weekends and holidays, and the heating demand increases accordingly. In addition, the temperature is controlled in zones according to the functions and personnel densities of different floors and areas, and the heating consumption needs to be adaptively adjusted. For the heating objects in hospitals, different departments have different requirements for temperature and humidity. For schools, heating has obvious seasonality and time periods. During the winter class period, areas such as classrooms and offices need heating to ensure a comfortable learning and working environment for teachers and students. During the winter and summer vacations, the heating demand drops significantly, and the school will reasonably adjust the operation of the heating system according to the course schedule and work and rest time to save energy. Therefore, for different secondary network heating objects, it is necessary to establish portraits of each heat user in the secondary network, set different heating operation modes according to different portraits of secondary network heat users, and provide personalized time-division, zone-based, and household-based control strategies.

[0082] In this embodiment, the prompt words for constructing portraits of each heat user in the secondary network include:

[0083] Role setting: Clearly define that the large model is an expert in the heating field during this conversation process, and guide the large model to output the behavior patterns and response methods of heat user portraits;

[0084] Requirement description: Describe the task requirements completed by the large model, including feature extraction and feature importance analysis;

[0085] Definition of multiple portrait dimensions: Include the definition of user basic characteristics and the definition of user heating behaviors, guiding the large model to correctly understand specific meanings. Among them, the definition of user basic characteristics is the static information of heat users; the definition of user heating behaviors is the dynamic information of heat users, including heat user work and rest behavior information, heat user heat consumption preference information for each time period, flexible response information, operation mode preference information, and heating feedback information.

[0086] It should be noted that the daily routine behavior information of heat users refers to relevant information such as the daily living and activity time patterns of heat users. For example, office workers leave early and return late, and are at home less during the day; the elderly are used to going to bed early and getting up early, and are active at home in the early morning and at night. Such information can help the heat supply side understand the time periods when users are at home, so as to adjust the heat supply strategy accordingly. For example, the temperature can be raised in advance before the user arrives home and the heat supply intensity can be reduced when the user leaves home;

[0087] The heat consumption preference information of heat users in each time period refers to the heat demand tendency of heat users at different time periods. For example, some families like a slightly lower temperature when resting at night, and hope the indoor temperature can rise rapidly for a period of time after getting up in the morning; for commercial places such as shopping malls, the heat consumption preferences are different during different time periods such as before business, during business, and after business ends, due to different numbers of people and activity situations. Mastering such information helps the heat supply system allocate heat more accurately and avoid energy waste;

[0088] The flexible response information refers to the degree of cooperation and response ability of heat users to the adjustment of the heat supply system. For example, when the heat supply side needs users to adjust their heat consumption behaviors due to special circumstances (such as equipment maintenance, energy allocation), whether users can respond in a timely manner and adjust the heat consumption equipment as required. At the same time, it also includes the willingness and ability of heat users to participate in demand-side response activities, such as reducing the heating load during peak periods;

[0089] The operation mode preference information refers to the preference and selection tendency of heat users for different heat supply operation modes. Common operation modes include energy-saving mode, conventional mode, adaptive mode, etc. Some users pay more attention to energy conservation and tend to choose the energy-saving mode; some users pursue a comfortable indoor environment and prefer the conventional heat consumption mode. By understanding the operation mode preferences of users, heat supply enterprises can provide heat supply services that better meet the needs of users and improve user satisfaction;

[0090] The heat consumption feedback information refers to the evaluations, opinions, and suggestions of heat users on heat supply services. For example, users feedback problems such as too high or too low indoor temperature, unstable heat supply, and equipment failures; it may also be the evaluation of the quality of heat supply services, such as the response speed of customer service and the service attitude of maintenance personnel. After the heat supply side collects such feedback information, it can promptly improve the heat supply system and services, and enhance the heat supply quality and management level.

[0091] In this embodiment, the heat user portraits of each secondary network are defined as:

[0092] P i =LLMs(D i ,R);

[0093] D i is the structured data set of the i-th heat user; R is the prompt word when constructing the heat user portrait; P iis the portrait of the i-th hot user; LLMs() is the large model used for constructing the portrait of hot users.

[0094] In this embodiment, in S2, the specific operation modes of each household in the secondary network of the heating system include:

[0095] Conventional operation mode: At any time during the heating season, the indoor temperature of the user reaches the preset temperature;

[0096] Energy-saving operation mode: It has a time period division mechanism and a customized temperature selection mechanism. It can select to save energy by setting a lower indoor temperature than the conventional preset temperature during a fixed time period according to the user's heat consumption behavior, and at the same time, economic compensation is given;

[0097] Intelligent adaptive mode: It can monitor the indoor temperature, the room activity time, and the number of flowing people in real time, and intelligently and automatically adjust the heat supply;

[0098] Economic optimization mode: According to the energy price fluctuation information, on the premise of meeting the basic comfort level, it provides the optimal operation plan to reduce the heating cost.

[0099] In actual application, the characteristics and applicable objects of the operation modes of each household in the secondary network of the heating system are introduced as follows:

[0100] 1) Conventional operation mode

[0101] Characteristics: During the entire heating season, at any time, the system will maintain the indoor temperature of the user at the preset temperature value to ensure that the indoor is warm and comfortable. This mode does not consider the changes in user behavior, energy price and other factors, provides continuous and stable heating, and is convenient for users with high requirements for heating stability and who do not need to participate in temperature regulation too much;

[0102] Heat consumption objects: Suitable for users with high requirements for indoor temperature stability and who do not care much about the changes in heating costs, such as places sensitive to environmental temperature like hospitals and nursing homes, and some residential users with high requirements for comfort and good economic conditions;

[0103] 2) Energy-saving operation mode

[0104] Characteristics: Through the time period division mechanism, a day or the entire heating season is divided into different time periods, such as day and night, weekdays and rest days, etc. At the same time, users can select a customized temperature lower than the conventional preset temperature during a fixed time period according to their own heat consumption behavior, so as to reduce heat consumption and achieve the purpose of energy saving. To encourage users to adopt this mode, economic compensation will also be given, such as fee reduction, point reward, etc.;

[0105] Heat users: Suitable for users who have a certain awareness of energy conservation, have regular work and rest schedules, and can adjust their temperature requirements according to time periods. For example, office workers can set a lower temperature when there is no one at home or less activity during the day; and industrial and commercial users who focus on cost control, such as warehouses, do not have particularly high temperature requirements;

[0106] 3) Intelligent adaptive mode

[0107] Features: With the help of sensors, the system monitors the indoor temperature, the time when people are active in the room, the number of mobile personnel and other information in real time. Based on this data, the system can intelligently and automatically adjust the heating supply. For example, when no one is in the room for a long time, the heating supply is automatically reduced; when there are more people or more activities, the heating supply is appropriately increased, thus ensuring comfort while saving energy to the greatest extent.

[0108] Heat users: Suitable for places with high requirements for comfort and energy saving, and where the activities of indoor personnel change frequently.

[0109] Such as office buildings, shopping malls and other commercial places;

[0110] 4) Economic optimization model

[0111] Features: This mode obtains information on energy price fluctuations in real time, and calculates the optimal operation plan through algorithms based on the premise of meeting basic comfort. For example, when energy prices are low, the heating intensity is appropriately increased to store some heat; when energy prices are high, the heating intensity is reduced and the stored heat is used to maintain the indoor temperature, thereby effectively reducing heating costs;

[0112] Heat users: Mainly suitable for industrial and commercial users who are more sensitive to heating costs, such as factories, large office buildings, etc. These places consume a lot of heat and the heating costs account for a relatively high proportion. Usually, some of them are equipped with independent energy supply equipment (such as air source heat pumps, ground source heat pumps, photovoltaic heating and other equipment), which can be autonomously controlled. This mode can significantly reduce operating costs.

[0113] In this embodiment, S2 specifically includes:

[0114] Obtain the definition of each household operation mode of the secondary network of the heating system, the heat user information of each household operation mode selected historically, and the operation parameters of the operation mode;

[0115] Set prompt words to guide the large model to generate the operation mode portrait, and clarify its function in generating the description of the operation mode portrait;

[0116] The name of the operation mode, the description of the mode characteristics, the association information between each operation mode and the hot user portrait, and the operation parameters of the operation mode are used as the operation mode data and input into the big model in combination with the prompt words to generate the operation mode portrait, which is expressed as:

[0117] P j = LLMs′(L j , M);

[0118] L j is the operation mode data of the j-th operation mode; M is the prompt word when generating the operation mode portrait; P j is the portrait of the j-th operation mode; LLMs′ is the large model used when generating the operation mode portrait.

[0119] In this embodiment, the large model generating the operation mode portrait includes:

[0120] Using a tokenizer to split the input sequence data into tokens, and mapping each token to a low-dimensional vector space to obtain token embeddings. At the same time, adding position encoding allows the large model to learn the position information of tokens in the sequence;

[0121] Based on the multi-layer transformer encoder structure of the large model, using the multi-head self-attention mechanism of each encoder enables the large model to learn the relationships between tokens in different representation subspaces, capture semantic information, and then using a feed-forward neural network to perform information transformation and feature extraction on the output of the multi-head self-attention, learning different abstract representations of the input data at different levels. By continuously stacking, a feature representation containing rich information of the input data is obtained;

[0122] At the output layer, using a linear transformation to map this feature representation to a dimension with the same size as the vocabulary size to obtain the prediction score of each token; using a decoding strategy to select the token with the highest probability according to the prediction score as the next generated token, and gradually generating the text content of the operation mode portrait.

[0123] In practical applications, select a large language model suitable for processing the fusion of structured and unstructured data, adapt the model so that it can understand the professional terms and data structures in the heating field, and then fine-tune it to enhance the understanding ability of heating-related knowledge; organize the household operation mode information and associated heat user profile data into a format suitable for input to the large model. Input the organized data into the large model, and use specific prompt words to guide the model to generate an operation mode profile. The prompt words can be designed as: "Based on the following household operation mode and related heat user profile information, generate a detailed profile of this operation mode, including but not limited to the main features, applicable scenarios, user group characteristics, energy efficiency performance, etc. of this mode." The model generates a profile description of the household operation mode according to the input data and prompt words. Manually review and optimize the operation mode profile results generated by the large model. Check whether the generated profile accurately reflects the characteristics of the operation mode and the user group characteristics. If there are inaccuracies or incompleteness, adjust the input data or prompt words and run the model again until a satisfactory operation mode profile is generated. At the same time, it can be verified by comparing with actual operation data and user feedback to ensure the authenticity and reliability of the profile.

[0124] It should be noted that for tokenization: use a tokenizer to split the input sequence into individual tokens. For example, the "energy-saving operation mode" may be tokenized into tokens such as "energy-saving", "operation", and "mode"; Embedding layer: map each token to a low-dimensional vector space to obtain token embeddings. At the same time, in order for the model to learn the position information of the tokens in the sequence, positional encoding is also added. The positional encoding can be fixed sine-cosine function encoding or learnable parameters. Finally, each token in the input sequence corresponds to an embedding vector containing token information and position information.

[0125] In the multi-head self-attention mechanism, for attention calculation: in each encoder layer, first perform multi-head self-attention calculation. The multi-head self-attention mechanism passes the input embedding vectors through multiple linear transformations respectively to obtain three matrices of query, key, and value. Then, calculate the similarity score between the query and the key, usually using the dot product operation, and then convert the score into a probability distribution through the softmax function. Finally, perform weighted summation on the values according to this probability distribution to obtain the attention output; Multi-head combination: concatenate the attention outputs of multiple heads together, and then perform dimensionality reduction through a linear transformation to obtain the final multi-head self-attention output. The multi-head self-attention mechanism enables the model to learn the relationships between tokens in different representation subspaces and capture richer semantic information.

[0126] The output of the multi-head self-attention is fed into a feed-forward neural network. The feed-forward neural network usually consists of two fully-connected layers, with a non-linear transformation using activation functions such as ReLU in the middle; the feed-forward neural network further performs information transformation and feature extraction on the output of the multi-head self-attention, enhancing the model's expressive ability. Additionally, residual connections are used on the outputs of both the multi-head self-attention and the feed-forward neural network, that is, the input is directly added to the output. Residual connections can alleviate the problem of vanishing gradients and make the model easier to train.

[0127] The above multi-head self-attention, feed-forward neural network, and residual connection are combined to form an encoder layer, and then multiple encoder layers are stacked together to form a multi-layer Transformer encoder. Each encoder layer learns a more advanced abstract representation based on the previous layer. Through continuous stacking, the model can gradually capture the complex semantic and structural information of the input data.

[0128] As Figure 3 shown, in this embodiment, S3 specifically includes:

[0129] Integrate the heat user portraits and operation mode portraits of the secondary network into an input data set;

[0130] Set prompt words to guide the large model to output personalized operation mode recommendation results and recommendation scripts, and clarify the role of the large model, task requirements description, and thinking chain construction during operation mode recommendation and script recommendation;

[0131] Input the input data set and prompt words into the large model. The large model uses the internal attention mechanism and neural network layers to calculate the similarity between the heat user portrait features and the features of each operation mode portrait;

[0132] Based on the similarity calculation results, the large model selects the operation mode with the highest similarity as the personalized recommendation result for the heat user to output;

[0133] Based on the output personalized operation mode recommendation result, the large model combines the pre-established script corpus to output differentiated recommendation scripts for different heat users.

[0134] It should be noted that when performing operation mode recommendation and script recommendation, define the role of the large model: explain the responsibilities of this role to the large model, such as "As an intelligent heating consultant, you need to provide the most suitable heating operation mode recommendation based on the detailed information of the heat user, and give the recommendation reasons and usage suggestions in an easy-to-understand and friendly language". Let the large model clearly understand its own position and the goals to be achieved.

[0135] Task requirement description, including input data description: Provide detailed information about the data input to the large model, such as "You will receive the rest behavior information of heat users, heat consumption preference information for each period, flexible response information, operation mode preference information, and heat consumption feedback information", so that the large model can understand the data range it can rely on. Recommended result requirements: Clearly define the presentation form of the operation mode recommendation result, for example, "The recommended result should include at least two most suitable operation modes and be sorted in descending order of matching degree". At the same time, stipulate the style and content key points of the recommended speech, "The recommended speech should be concise and clear, highlighting how each operation mode meets the specific needs of users, and the length of the speech is controlled between 100 - 150 words".

[0136] Thought chain construction: Guide the large model to analyze according to certain logical steps, for example, "First, analyze the rest behavior information of heat users to judge the time distribution law of users at home, and initially screen out the possibly suitable operation modes. Then, combine the heat consumption preference information for each period to further determine the energy-saving or comfort advantages of each mode in different periods. Refer to the flexible response information to evaluate the acceptance degree of users for different adjustment methods and adjust the recommended modes"; finally, output the recommended speech according to the recommended operation mode.

[0137] As Figure 4 shown, in this embodiment, S4 specifically includes:

[0138] Obtain the heat metering data of each heat user, including heat consumption, flow rate, and temperature in different time periods;

[0139] Obtain the household control data of each heat user, including the operation mode switching time and the adjustment parameters of the control equipment;

[0140] Obtain the control knowledge corresponding to the operation mode, including: the control mechanism of the control equipment and the target temperature range under different operation modes;

[0141] According to the household heat metering data, control data, and control knowledge corresponding to the operation mode of each heat user, use deep learning methods to perform entity, relationship, and attribute extraction respectively; the relationship extraction includes the relationship between heat users and control equipment, the relationship between heat users and operation modes, and the relationship between control equipment and operation modes;

[0142] Build a household control knowledge graph according to the extracted entities, relationships, and attributes, including: determine the structure of the secondary network household control knowledge graph, in the form of a directed graph, with nodes representing entities and edges representing the relationships between entities; add entities as nodes to the knowledge graph, create corresponding edges according to the relationships, and add corresponding attribute information to each node and edge.

[0143] It should be noted that the entities include heat user entities, control equipment entities, and operation mode entities; the attribute extraction includes: for heat user entities, extracting heat consumption, temperature, etc. in their heat metering data as attributes; for control equipment entities, extracting opening degree, adjustment time, etc. in their control data as attributes; for operation mode entities, extracting target temperature range, energy saving rate, etc. in the control knowledge as attributes.

[0144] As Figure 5 shown, in this embodiment, the step S5 specifically includes:

[0145] Associating and integrating the operation mode data selected by the heat user with the heat user portrait data;

[0146] Extracting knowledge information related to the heat user operation mode and control equipment from the household control knowledge graph, and converting it into a vector form suitable for model input, and fusing the processed knowledge graph vector with the heat user data;

[0147] Converting the heat network hydraulic balance mechanism knowledge into a mathematical model and constraint conditions, and extracting the parameters related to the heat network hydraulic balance;

[0148] Defining the training objective: aiming to generate the optimal secondary network household control strategy, setting the training objective function according to the operation mode requirements of the heat user, minimizing the heat network energy consumption, and ensuring the heat network hydraulic balance;

[0149] Integrating the associated data of the operation mode and the heat user portrait, the knowledge graph data, and the heat network hydraulic balance data as training samples and inputting them into the large model for training to generate a secondary network household control large model for the heating system, and continuously adjusting the parameters of the large model so that the large model can learn the relationship between the heat user operation mode, the heat user portrait, the household control knowledge graph, and the heat network hydraulic balance;

[0150] For the large model after training and optimization, inputting the new heat user operation mode, heat user portrait, and real-time operation data of the secondary network, the large model generates a secondary network household control strategy according to the learned knowledge and relationships, including the adjustment parameters of the pre-household control equipment and the switching time of the operation mode.

[0151] In practical applications, the training and optimization of the large model for household-by-household regulation in the secondary network of the heating system adopt a deep reinforcement learning algorithm, specifically including: representing the environmental state of household-by-household regulation in the secondary network through the operating mode selected by heat users, heat user portraits, the knowledge graph of household-by-household regulation, and the hydraulic balance condition of the heat network where the heat user buildings are located; taking the regulation parameters of the regulation equipment in front of the households in the secondary network as the action space, and designing a reward mechanism based on the regulation effect of the action; the regulation effect is preset to minimize the heat network energy consumption and ensure the hydraulic balance of the heat network on the premise of meeting the operating modes of each heat user; by continuously performing actions, observing results, obtaining rewards, and updating the value function Q value of each action, the large model can learn the optimal household-by-household regulation strategy for the secondary network.

[0152] In several embodiments provided in the present 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 show the possible architectures, functions, and operations of systems, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code contains one or more executable instructions for implementing the 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 from that marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0153] In addition, each functional module in various embodiments of the present invention can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part. If the function is implemented in the form of 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 the present invention, in essence, or the part that contributes to the prior art, or a part of this 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 for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories, random access memories, magnetic disks, or optical discs that can store program codes.

[0154] Inspired by the above-described ideal embodiments of the present invention, through the above description, relevant staff can completely make various changes and modifications without departing from the technical idea of this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A personalized regulation method for the secondary heating network based on large models and multimodal portraits, characterized in that, It includes: S1. Obtain the multimodal data of each heat user in the secondary network of the heating system, and after performing feature extraction and importance analysis using a large model, construct user portraits of each heat user in the secondary network; S2. Extract information on the operation modes of each household in the secondary network of the heating system, combine with the user portraits of the heat users who select the operation modes of each household, and use a large model to generate operation mode portraits of the operation modes of each household; S3. Input the user portraits and operation mode portraits of each heat user in the secondary network into the large model for matching analysis, and output personalized operation mode recommendation results and recommendation scripts for each heat user; S4. Establish a household control knowledge graph based on the household heat metering data, regulation data of each heat user, and regulation knowledge corresponding to the operation modes; S5. According to the operation mode recommendation results and recommendation scripts of each heat user, recommend to the heat users to select the operation mode. Then, based on the operation mode selected by the heat user, the user portrait, and the household control knowledge graph, combined with the knowledge of the heat network hydraulic balance mechanism, train and optimize the household control large model of the secondary network of the heating system, and output the household control strategy of the secondary network.

2. The personalized regulation method for the secondary heating network according to claim 1, wherein The specific content of S1 includes: Obtain the multimodal data of each heat user in the secondary network of the heating system, including text data, voice data, and image data; the text data includes basic information such as heating user type, user age, house orientation, house area, floor where the user is located, as well as user's work and rest time, heat consumption in each period, flexible heat load response information, complaint information, maintenance work order information, and preferred operation mode; the voice data includes voice evaluation information on heating effect, fault description voice information, and special heat use requirement voice information between the heat user and the heating enterprise; the image data includes images of the indoor heating equipment status, outdoor heating pipeline images, house exterior images, house layout images, and images of people flowing in the house; the heating user type includes civil heating and industrial and commercial heating; Preprocess and perform multimodal fusion on the obtained multimodal data to form a structured data set, and input it into a pre-trained large model. Use the method of combining the large model and prompt words. Take the structured data set and prompt words as inputs, and then use the attention mechanism to perform feature extraction on the data, including feature extraction of text, voice, and image, to obtain the basic features of the heat user and the heat use behavior features of the user. After performing feature importance analysis using the feature importance evaluation method, output user portraits of each heat user in the secondary network including multiple portrait dimensions.

3. The personalized regulation method for the secondary heating network according to claim 2, wherein The prompt words when constructing the user portraits of each heat user in the secondary network include: Role setting: Clearly define that the large model is an expert in the heating field during this conversation process, and guide the large model to output the behavior patterns and response methods of the user portraits; Requirement description: Describe the task requirements completed by the large model, including feature extraction and feature importance analysis; Multi-image dimension definition: It includes the definition of user basic characteristics and the definition of user heat usage behavior to guide the large model to correctly understand specific meanings. Among them, the definition of user basic characteristics is the static information of hot users, and the definition of user heat usage behavior is the dynamic information of hot users, including hot user's daily routine behavior information, heat consumption preference information at each time period, flexible response information, operation mode preference information, and heat usage feedback information.

4. The personalized regulation method for the secondary heating network according to claim 2, wherein, The definition of each hot user portrait in the secondary network is: P i = LLMs(D i ,R); D i is the structured data set for the i-th hot user; R is the prompt word when constructing the hot user portrait; P i is the portrait of the i-th hot user; LLMs() is the large model used when constructing the hot user portrait.

5. The personalized regulation method for the secondary heating network according to claim 1, wherein In S2, the specific operation modes of each household in the secondary network of the heating system include: Conventional operation mode: At any time during the heating season, the indoor temperature of the user reaches the preset temperature. Energy-saving operation mode: It has a time period division mechanism and a customized temperature selection mechanism. It can select to save energy by setting a lower indoor temperature than the conventional preset temperature during a fixed time period according to the user's heat usage behavior, and at the same time give economic compensation. Intelligent adaptive mode: Real-time monitor the indoor temperature, room activity time, and the number of flowing people, and intelligently and automatically adjust the heat supply. Economic optimization mode: According to the energy price fluctuation information, provide the optimal operation plan under the premise of meeting the basic comfort level to reduce the heating cost.

6. The personalized regulation method for the secondary heating network according to claims 1 and 5, characterized in that, S2 specifically includes: Obtain the definition of each household operation mode in the secondary network of the heating system, the hot user information of each household operation mode selected in history, and the operation parameters of the operation mode. Set the prompt words to guide the large model to generate the operation mode portrait, and clarify its function in the description of the operation mode portrait generation. Take the name of the operation mode, the description of the mode characteristics, the association information between each operation mode and the hot user portrait, and the operation parameters of the operation mode as the operation mode data, and input them into the large model together with the prompt words to generate the operation mode portrait, expressed as: P j = LLMs′(L j , M); L j is the operation mode data for the j-th operation mode; M is the prompt word when generating the operation mode portrait; P j is the portrait of the j-th operation mode; LLMs′ is the large model used when generating the operation mode portrait.

7. The personalized regulation method for the secondary heating network according to claim 6, characterized in that The generation of the operation mode portrait by the large model includes: Use the tokenizer to split the input sequence data into tokens, and map each token to a low-dimensional vector space to obtain token embeddings. At the same time, add position encoding to let the large model learn the position information of the tokens in the sequence. Based on the multi-layer transformer encoder structure of the large model, use the multi-head self-attention mechanism of each encoder to enable the large model to learn the relationship between tokens in different representation subspaces, capture semantic information, and then use the feed-forward neural network to perform information transformation and feature extraction on the output of the multi-head self-attention, and learn different abstract representations of the input data at different levels. By continuously stacking, obtain the feature representation containing rich information of the input data. At the output layer, use a linear transformation to map this feature representation to a dimension with the same size as the vocabulary size to obtain the prediction score of each token; use the decoding strategy to select the token with the highest probability according to the prediction score as the next generated token, and gradually generate the text content of the operation mode portrait.

8. The personalized regulation method for the secondary heating network according to claim 1, wherein S3 specifically includes: Integrate the hot user portraits of each household in the secondary network and the operation mode portraits into an input data set. Set the prompt words to guide the large model to output personalized operation mode recommendation results and recommendation scripts, and clarify the role of the large model, task requirement description, and thinking chain construction during operation mode recommendation and script recommendation. Input the input dataset and the prompt words into the large model, and the large model uses the internal attention mechanism and neural network layers to calculate the similarity between the hot user portrait features and the portrait features of each operation mode; Based on the similarity calculation results, the large model selects the operation mode with the highest similarity as the recommended result of hot user personalization and outputs it; Based on the output personalized operation mode recommendation result, the large model combines the pre-established speech corpus to output differentiated recommendation speech for different hot users.

9. The personalized regulation method for the secondary heating network according to claim 1, characterized in that The specific steps of S4 are as follows: Obtain the heat metering data of each hot user, including heat consumption, flow rate, and temperature in different time periods; Obtain the household control data of each hot user, including the operation mode switching time and the adjustment parameters of the control equipment; Obtain the control knowledge corresponding to the operation mode, including: the control mechanism of the control equipment and the target temperature range under different operation modes; Based on the household heat metering data, control data, and control knowledge corresponding to the operation mode of each hot user, use deep learning methods to extract entities, relationships, and attributes respectively; the relationship extraction includes the relationship between the hot user and the control equipment, the relationship between the hot user and the operation mode, and the relationship between the control equipment and the operation mode; Build a household control knowledge graph according to the extracted entities, relationships, and attributes, including: determining the structure of the secondary network household control knowledge graph, using a directed graph form, representing entities with nodes and relationships between entities with edges; adding entities as nodes to the knowledge graph, creating corresponding edges according to the relationships, and adding corresponding attribute information to each node and edge.

10. The personalized regulation method for the secondary heating network according to claim 1, wherein The specific steps of S5 are as follows: Associate and integrate the operation mode data selected by the hot user with the hot user portrait data; Extract the knowledge information related to the hot user's operation mode and control equipment from the household control knowledge graph, and transform it into a vector form suitable for model input, and fuse the processed knowledge graph vector with the hot user data; Transform the knowledge of the heat network hydraulic balance mechanism into a mathematical model and constraint conditions, and extract the parameters related to the heat network hydraulic balance; Define the training objective: aiming to generate the optimal secondary network household control strategy, set the training objective function according to the operation mode requirements of the hot user, minimize the heat network energy consumption, and ensure the heat network hydraulic balance; Integrate the associated data of the operation mode and the hot user portrait, the knowledge graph data, and the heat network hydraulic balance data as training samples and input them into the large model for training to generate a large model for secondary network household control of the heating system, and continuously adjust the parameters of the large model so that the large model can learn the relationships between the hot user operation mode, the hot user portrait, the household control knowledge graph, and the heat network hydraulic balance; For the large model after training and optimization, input the new hot user operation mode, hot user portrait, and real-time operation data of the secondary network. The large model generates a secondary network household control strategy according to the learned knowledge and relationships, including the adjustment parameters of the pre-household control equipment and the switching time of the operation mode.

Images