Water supply temperature intelligent regulation and control method based on thermal characteristic analysis and migration correction

Abstract

The invention relates to a water supply temperature intelligent regulation and control method based on thermal characteristic analysis and migration correction. The method comprises the following steps: S1, screening thermal engineering principle characteristics; s2, data acquisition and processing; s3, segmenting the data set; s4, determining a basic model; s5, performing thermal feature recognition; s6, model combination; and S7, model application. According to the method, the load predicted value and the hydraulic flow serve as core input, the water supply temperature is expressed as a simple function form, and key thermal characteristics of pipe network characteristics in historical data are mined and transferred to a real-time operation state to be corrected and optimized. According to the method, a thermal mechanism architecture is dominated, uncertainty factors are represented, an intelligent algorithm is introduced, modular design is adopted, a lightweight modeling scheme is formed, robustness can be kept under data fluctuation and working condition changes, interpretability, self-learning, self-adaption and continuous optimization capacity are achieved, and the method is suitable for large-scale popularization and application. Therefore, dynamic prediction and quick response of the water supply temperature are realized.

Description

Technical Field

[0001] This invention belongs to the field of centralized heating technology, specifically relating to an intelligent water supply temperature control method based on thermal feature analysis and migration correction. Background Technology

[0002] Centralized heating systems are widely used in building and industrial heating in northern my country. "On-demand heating" is the fundamental approach to ensuring users' heat needs while achieving significant energy conservation and emission reduction. However, due to significant system thermal inertia, building thermal inertia, and the time-varying nature of user heating patterns, centralized heating systems have limited transient response capabilities, resulting in varying degrees of supply-demand mismatch and energy waste in actual heating processes. With the continuous expansion of centralized heating systems in northern my country and the increasing demands for user comfort, intelligent and energy-efficient control of heating system operation has become an important direction for the transformation of urban energy systems.

[0003] Heat load forecasting is often considered a key element in accurate heating management and plays an important role in evaluating the most effective energy-saving strategies [1]. Recent advances in Internet of Things (IoT) technology and automated control systems have enhanced the intelligence of heating systems, enabling more precise regulation to meet on-demand heating needs [2]. Numerous studies have improved the accuracy of heat load forecasting models [3-4]. Even with accurate load forecasting, many current heating systems still face challenges in selecting appropriate regulation strategies to match fluctuating load demands. This often leads to energy inefficiencies due to over-reliance on manual control [5]. Accurate load forecasting is crucial for guiding the operation of heating systems; therefore, to achieve intelligent heating regulation, load forecasting results need to be decomposed into appropriate control parameters.

[0004] The main purpose of centralized heating system operation regulation is to ensure that the heating system meets users' heat demand while avoiding energy waste caused by overheating. Many scholars have conducted research on the operation regulation of centralized heating systems. Specific operation regulation methods can be broadly categorized into three types: quantitative regulation that only changes the system flow rate, qualitative regulation that only changes the supply water temperature, and qualitative regulation based on staged flow rate regulation. For example, in energy stations with relatively large loads on the heat source side, a large number of heating users, and significant load fluctuations throughout the heating season, qualitative regulation based on staged flow rate regulation is typically employed. In this process, the supply water temperature, as the core parameter for controlling heat transfer efficiency and the most basic data recorded in daily operation and maintenance, directly determines user end-point thermal comfort, network heat transmission efficiency, and heat source operating energy consumption. Regarding the operation regulation of centralized heating systems, many studies have shown that the operation regulation strategy of centralized heating systems is not directly related to the building's design heat load index, and there is a significant difference between the actual and design parameters. Therefore, the actual operation regulation scheme of a centralized heating system differs from the theoretical operation regulation scheme, and the operation regulation scheme needs to be formulated based on the actual parameters.

[0005] In the actual operation of centralized heating systems, traditional practices are mostly based on historical operating experience and determined by future changes in outdoor temperature. For example, the water supply temperature is set according to the outdoor temperature using an empirical climate compensation curve. Although this method is simple in structure and easy to implement, it does not take into account dynamic factors such as building thermal inertia, meteorological disturbance coupling, actual heat load changes and user-end temperature feedback. This can easily lead to the heating system being "overheated" or "underheated", resulting in serious energy waste and difficulty in ensuring comfort.

[0006] Chen Tingting et al. [6] proposed a study on the prediction of heating load of existing buildings and a climate compensation model based on multiple meteorological parameters. The model is mainly based on multiple meteorological parameters such as outdoor temperature, wind speed and humidity. The water supply temperature setting curve is optimized by multiple regression, and a static function fitting model is constructed. This method shows good accuracy in historical data fitting (R²). 2 The error is reduced to 0.92, which is about 15% lower than that of the univariate external temperature model. However, this type of method is only applicable to static meteorological scenarios and cannot adapt to dynamic heat load changes.

[0007] Zhang Zhoukang et al. [7] proposed a multi-strategy intelligent heating method based on temperature and time offset control. The water supply temperature setting method proposed in this patent sets the secondary network water supply temperature control curve according to the building type and outdoor temperature, collects the ambient temperature in real time and adjusts the opening of the primary side electric valve through PID algorithm to achieve indirect control of the secondary network temperature; for the primary network direct supply system, the boiler load is adjusted within the set temperature range by setting the "boiler water supply temperature - outdoor temperature" relationship curve and combining fuzzy control strategy to avoid frequent fluctuations. The system also sets a reasonable operating range for the regulating valve (20%–80%), and when the opening exceeds the limit, the frequency of the primary network circulating pump is adjusted in linkage to ensure water supply flow matching. In addition, the method introduces time offset logic control, dynamically offsets the water supply temperature according to the user demand in different time periods, and further realizes the unity of energy-saving operation and heating comfort. However, the overall system still belongs to rule-driven control logic, and the response flexibility is limited when facing complex nonlinear load changes, building thermal inertia differences and sudden weather events.

[0008] Wang Jun et al. [8] proposed a method for constructing a two-network water supply prediction model based on piecewise linear functions. This method comprehensively considers four types of variables: outdoor air temperature, light intensity, time factor, and temperature rise and fall demand. It establishes a monotonic model for outdoor air temperature and water supply temperature, a penalty coefficient for light intensity, time-sharing water supply temperature strategy variables, and a temperature rise and fall correction model, and integrates them into a unified water supply temperature prediction formula. While maintaining monotonicity, the model can dynamically respond to changes in the external environment and ensure that the heating system adjusts the water supply temperature as needed. However, the setting of multiple parameters is empirical and lacks theoretical basis, which limits the universality and transferability of the model.

[0009] Feng Encheng et al. [9] proposed a model-based method for precise on-demand control of centralized heating systems. Based on historical operating data and weather data of the heating station, a load prediction model for the heating station is established. Under given weather conditions, the short-term heat load of the heating station for the next day is predicted. Based on the required heat load, the required secondary side supply temperature and flow rate are determined. Meteorological parameters, supply water temperature, return water temperature and flow rate are input. A heat load prediction model is established using a data mining algorithm (taking the SVR algorithm as an example). The return water temperature at the current moment is used as the calculated return water temperature. The secondary side supply water temperature that meets the predicted heating load demand is calculated according to the heat transfer formula.

[0010] In order to improve the adaptability and precision of control, researchers have begun to introduce control methods based on room temperature feedback. Wang Yanmin et al.

[10] proposed an on-demand control strategy with "effective room temperature" as the core indicator, and dynamically corrected the water supply temperature setting through user-side thermal feedback. However, such methods are limited in practical applications due to factors such as indoor temperature collection coverage and the complexity of IoT deployment, and are difficult to promote in large-scale heating areas. At the same time, combined with predictive feedback to form a joint control system, the water supply temperature adjustment cannot be modularized and cannot be used independently.

[0011] Luo et al.

[11] proposed a method and device for determining the water supply temperature of a heating station. This control method obtains the real indoor temperature and outdoor ambient temperature of the user in real time through sensors, constructs a prediction model using historical data, and predicts the indoor temperature of the user space based on the current outdoor temperature. When the difference between the predicted value and the actual room temperature exceeds the set threshold, the system determines that the current heat supply does not match the heat demand of the user, and automatically starts the adjustment mechanism. Based on the predicted room temperature and room thermal parameters (such as volume, heat loss, radiator performance), the required water supply temperature of the heat dissipation equipment is first determined, and then the heat transfer path and temperature attenuation between the heat dissipation equipment and the heating station are combined to back-calculate the target water supply temperature that the heating station should output. This method takes "single target space" as the basic control object, which is difficult to uniformly control in a complex heating network with multiple users and multiple buildings connected at the same time; at the same time, the room temperature prediction model is trained based on historical data, and the model accuracy may decrease under the circumstances of drastic climate change, building structure change, sudden change in user behavior, etc.

[0012] Liang Xue et al.

[12] proposed a digital simulation system and method for heating temperature curve and hydraulic balance regulation. In this method, the water supply temperature setting is mainly based on the outdoor temperature. The outdoor temperature information is first collected by an external sensor, and the primary side water supply temperature is dynamically adjusted by controlling the electric regulating valve in the heat source simulation unit, thereby indirectly meeting the heating demand of the secondary network. The system supports the setting of eight or more outdoor temperature parameter points, and meets the temperature control requirements of different time periods through the curve translation function. During the regulation process, the optimal heating temperature curve under actual operation is determined by combining the feedback from the indoor temperature sensor and multiple experimental simulations. However, this method is only a simulation platform and is not a real-time online control.

[0013] Liu Guoqiang et al.

[13] proposed an online matching and regulation method for regional heating supply and demand. In this method, the water supply temperature is regulated by dynamically monitoring the operating status of the heating area weighted room temperature, electric valve opening, water supply flow, etc. of buildings in the area, and combined with the preset design parameters to establish a real-time regulation mechanism for water supply temperature and main pipe pressure difference. First, the building's design temperature, set temperature, radiator area, and other parameters are input, and the operating data is collected in real time. Then, based on the building's room temperature deviation, actual radiator configuration, and flow rate, multiple regulation parameters, including comprehensive room temperature correction factor, heat exchange factor, and flow rate regulation factor, are calculated and used to construct the regulation equation for the set water supply temperature and supply and return water pressure difference. Finally, by adjusting the primary side electric regulating valve and the secondary side circulating pump, efficient dynamic matching of heating supply and demand is achieved. Test results show that this method can increase the matching degree from about 70% in the traditional system to more than 90%. However, this method relies on a large amount of real-time data at the building level (such as electric valve opening, weighted room temperature, actual radiator area, etc.), which requires high coverage and accuracy of monitoring equipment and has a high implementation cost.

[0014] Guo Xiaojie et al.

[14] established a water supply temperature model based on the PSO-LSTM method. This method is based on the historical operation data of the centralized heating system and constructs a prediction model that integrates particle swarm optimization algorithm (PSO) and long short-term memory neural network (LSTM) for the characteristics of water supply temperature change. The key influencing factors (such as outdoor temperature, instantaneous heat flow, instantaneous water flow and water supply temperature in the previous few hours) that are highly correlated with water supply temperature are selected by Pearson coefficient analysis as model input. The model uses PSO to globally optimize the key parameters (including the number of neurons and learning rate) in the LSTM network to improve the prediction accuracy and convergence efficiency. The optimized model can achieve high-precision prediction of short-term water supply temperature and provide data support for the heating system to formulate a more scientific and reasonable water supply temperature setting strategy. The intelligent setting and accurate prediction of water supply temperature of the heating system are realized through the data-driven method, which improves the accuracy and practicality of prediction, but the data volume requirement is large.

[0015] Current methods for regulating water supply temperature can be mainly divided into four categories: static empirical models, combined heat load prediction models, data-driven prediction models, and rule-based control models. Static empirical models are suitable for macro-level planning and initial settings, but lack the ability to dynamically respond to operational disturbances. Combined modeling methods couple load prediction with temperature settings; the model accuracy depends on the synergistic effect between sub-modules, and small errors in preceding modules can be amplified layer by layer, affecting overall regulation performance. Data prediction models, while possessing high accuracy, have high requirements for the quality and quantity of historical data, making them difficult to promote in data-scarce or abnormal scenarios. Rule-based control models, although possessing strong engineering feasibility and simplicity, have limited adjustment flexibility and adaptive capabilities, making it difficult to meet the precise regulation needs under dynamic operating conditions.

[0016] In summary, the aforementioned water supply temperature models and regulation technologies all involve environmental factors such as outdoor air temperature and are correlated with demand load forecasting. Coupled with load forecasting and temperature setting, the model accuracy depends on the synergistic effect between sub-modules. Small errors in preceding modules can be amplified layer by layer, affecting overall regulation performance. In reality, the load demand characteristics exhibited by users at different levels of a centralized heating system vary significantly. Coupled with system thermal inertia, building thermal inertia, and the variability and time-varying nature of user demand, existing water supply temperature models and regulation technologies have limited engineering feasibility and are difficult to adapt to scenarios with varying levels of information and digitalization.

[0017] On the other hand, many studies have shown that "small flow rate and large temperature difference" is the economical operating mode for centralized heating, which to some extent indicates that the temperature difference between the supply and return water represents the system's operating level [15-16]. This mode is more suitable for relatively small-scale control units where minimum flow rate limits can be disregarded. In actual heating processes, a minimum flow rate is usually set to ensure hydraulic circulation for all users. Therefore, achieving target energy consumption control for on-demand heating through precise control of the supply water temperature based on the target load and set flow rate is of great significance for realizing clean and efficient heating in the building sector under the dual-carbon target.

[0018] As mentioned earlier, in the daily operation and maintenance of centralized heating systems, there is a common discrepancy between planned load and actual supply to varying degrees. Water supply temperature is one of the key monitoring parameters, directly affecting system energy consumption and frequently attracting attention from regulatory authorities and end-users. In the daily operation and regulation of district heating systems, energy stations often employ a phased flow regulation mode. It can be considered that flow rate is a constraint in the daily operation and maintenance scheduling of energy stations, and the "load output" is controlled by setting the "water supply temperature." Therefore, this invention proposes a novel water supply temperature model and dynamic regulation method based on target energy consumption management to achieve control over the "load output" on the heat source side of the energy station, more accurately matching the "demand load" on the user side, and minimizing energy waste caused by overheating.

[0019] References: [1] Yang Aixin. Building energy consumption transfer learning load prediction based on cloud-edge collaboration [J]. China Science and Technology Information, 2025, (10): 59-61. [2] Yan Xiuying, Men Qi, Wu Xiaoxue. Deep transfer learning method for short-term load forecasting across buildings [J]. Journal of Electric Power System and Automation, 2025, 37(04):88-97. [3] Han Yingjie, Xu Yuanyuan, Hu Rong. Research on heating load prediction under multi-parameter conditions based on PSO-SVM [J]. Science and Technology Innovation and Application, 2025, 15(08):73-76. [4] Zhang Changhao, Lu Xukun. Comparative study on neural network prediction models of heat load in heating stations [J]. Gas and Heat, 2024, 44(06):1-5. [5]Xiaojie Lin, Yihui Mao, Jiaying Chen, Wei Zhong. Dynamic modeling and uncertainty quantification of district heating systems consideringrenewable energy access[J].Applied Energy,2023,349:121629. [6] Chen Tingting, Han Yuhang, Yu Xi, et al. Research on heating load prediction of existing buildings and climate compensation model based on multiple meteorological parameters [J]. Energy Conservation, 2024, 43(03):13-16. [7] Zhang Zhoukang, Tang Xiaodong, Yang Qianwen. A multi-strategy intelligent heating method based on temperature and time offset control [P]. Shandong Province: CN112539450B, 2021-11-05. [8] Wang Jun, Ren Yumeng, Zhang Yinlu, et al. A method for constructing a prediction model for secondary water supply based on piecewise linear functions [P]. Tianjin: CN116680935B, 2023-10-13. [9] Feng Encheng, Lin Xiaojie, Huang Wei, Wang Liteng, Zhong Wei. A model-based method for precise on-demand control of centralized heating systems [P]. Zhejiang Province: CN111473407B, 2021-03-30.

[10] Wang Yanmin, Li Zhiwei, Liu Junjie, et al. Research on demand-based control technology of centralized heating based on room temperature monitoring feedback [J]. Heating Ventilating & Air Conditioning, 2024, 54(10):44-51.

[11] Luo Wen, Su Hui, Liu Yadan, et al. Methods, devices and systems for determining water supply temperature in heating stations [P]. Inner Mongolia: CN115899809B, 2025-05-16.

[12] Liang Xue, Zhao Zhiqiang, Pang Haicheng, et al. Digital simulation system and method for heating temperature curve and hydraulic balance regulation [P]. Shandong Province: CN112555979B, 2021-10-01.

[13] Liu Guoqiang, Ying Yuzheng, Yan Gang. Online matching and adjustment method for heat supply and heat demand in district heating system [P]. Shaanxi Province: CN113028493B, 2022-04-05.

[14] Guo Xiaojie, Ma Wenjing, Cao Shanshan, et al. Analysis of the influence of PSO-LSTM model on water supply temperature prediction [J]. Gas and Heat, 2025, 45(01):12-16.

[15] Zhang Bingli, Zhang Jingyun. The operation mode of "small flow rate and large temperature difference" can be realized [J]. District Heating, 2010, (04): 13-15.

[16] Zheng Lihong, Zhou Zhihua, Shi Wanliang. Research on operation and regulation model and evaluation index of heating system[J]. Building Energy Conservation (Chinese and English), 2024, 52(12):107-113. Summary of the Invention

[0020] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent water supply temperature control method based on thermal feature analysis and transfer correction. Using load forecasts and hydraulic flow rates as core inputs, the method expresses the water supply temperature as a concise function. By mining key thermal features reflecting system inertia and network characteristics from historical data, these features are transferred to the real-time operating state for correction and optimization, achieving dynamic prediction and rapid response of the water supply temperature. This invention is guided by a thermal mechanism architecture, characterizes uncertainties, introduces intelligent algorithms, and adopts a modular design to form a lightweight modeling scheme. It remains robust under data fluctuations and changes in operating conditions, and possesses interpretability, self-learning, self-adaptation, and continuous optimization capabilities.

[0021] The technical problem solved by this invention is achieved through the following technical solution: A smart water supply temperature control method based on thermal feature analysis and migration correction is proposed. This method combines fundamental principles of thermal engineering with a thermal feature recognition mechanism to establish a simplified yet efficient water supply temperature prediction and control model, enabling rapid modeling and adjustment based on two core quantities: load and flow rate. The control system employed by this method includes a peripheral module, a control module, and a controlled object. The peripheral module provides historical / real-time data input, the control module completes model construction and calibration, model application, and outputs adjustment commands, and the controlled object responds to the commands, executes operational adjustments, and generates operational demand feedback. Data acquisition in the peripheral module relies on flow sensors, heat meters, and water supply temperature sensors deployed at key nodes of the heating network. The data acquisition module obtains real-time operating data of the heating system, including water supply temperature. t g Heat load Q and traffic G The collected data is transmitted to a host computer and stored in a database, serving as the basis for modeling and prediction. The steps of the control method are as follows: S1. Feature selection based on thermal engineering principles: Starting from the basic heat balance principle of the heating system, we search for more universal input features and determine the input features with the goal of parameter simplification. S2. Data Acquisition and Processing: By reading the database in the external modules and based on the actual historical data of the heating network operation, a parameter set with a consistent time scale is formed, including water supply temperature. t g ,flow G and heat load Q ; S3. Determine the basic model: Use parameter set data for basic model parameter calibration. Combine multiple regression or ensemble learning methods to establish the functional relationship between water supply temperature, heat load, and flow rate, and determine the basic model. f 0: T_static=f 0(Q , G ); S4. Thermal Feature Identification: Input the parameter set into the basic model to obtain the simulated water temperature. T_static With respect to the actual water supply temperature t g The deviation is defined as the residual term characterizing the time-varying drift of the system's thermal characteristics, and a residual mapping model is established. f 1: ε = f 1( Q , G ); Introducing a multi-model fusion learning framework f 1. Conduct training and integration to achieve adaptive representation of uncertain disturbances and drift characteristics under working conditions, and construct a thermal feature recognition module; S5. Model Combination: Combining the basic model f 0 and residual mapping model f 1. Combining these elements, we obtain a water supply temperature prediction and control model: T_pred = T_static + ε = f 0( Q , G )+ f 1( Q , G ); S6. Model Application: Divided into two stages based on the actual situation during the heating season: Phase 1 – Initial commissioning phase of the new heating season: Based on the planned heat load and flow limit of the hydraulic regulation of the heating system set in the peripheral modules, the water supply temperature prediction and control model will output the water supply temperature setpoint of the control unit. T_set ; Phase ② – Entering the new heating season operation phase: Collect actual operating data from the past few days (more than 2 days) or several weeks, and adjust the setpoints based on actual load and flow. T_set Perform verification and reset ， By incorporating actual water supply temperature, a rolling window retraining and automatic parameter update strategy is adopted to identify and incorporate new thermal features, and a residual correction function is constructed. e = f 2 ( Q , G ), and use it for T_set Perform real-time correction and optimization to form T_set = f 0( Q , G )+f 1( Q , G )+ f 2 ( Q , G ); in, f 2 Characterization runtime relative to f The incremental correction term of 1 is used to compensate for additional drift and disturbance under the new operating conditions.

[0022] Moreover, S1 specifically refers to: Based on the fundamental heat balance principle of a heating system, under ideal and stable operating conditions (ignoring heat loss during the heat network transmission process), the heat output from the heat source should perfectly match the actual heat load demand at the user end. The heat conservation relationship is as follows: (1) (2) (3) (4) In the formula, Q 1. Q 2 and Q 3 represents the actual heat load of the building, the heat dissipation of the heat dissipation equipment, and the load on the energy station side, respectively; q The building heat load index is expressed in W / (m²). 3 ·℃); V is the external structural volume of the building, m 3 ; t in and t out Indoor and outdoor temperatures, respectively, in °C; K It is the heat transfer coefficient of the heat dissipation device, W / (m²). 2 ·℃); A The heat dissipation area of ​​the heat dissipation device is expressed in meters (m). 2 ; c The specific heat capacity of hot water is J / (kg·℃). G The energy station's flow rate is kg / h. t g and t h These are the supply water temperature and the return water temperature, respectively, in °C; t p The average temperature of the supply and return water is expressed by the following formula: t p = ( t g + t h) / 2; When the system is identified as a heating system, formulas (1) to (4) are simplified to: (5) To further explore the correlation between water supply temperature and other parameters and to find more general input characteristics, the Pearson correlation coefficient was used to analyze the correlation between water supply temperature and the parameters described in formula (5). With the goal of simplification, for return water temperature, the correlation coefficient between the supply and return water temperature difference and the load reached a high level, indicating that the load can also reflect the change in the supply and return water temperature difference. The water supply temperature is lightly described by the relationship presented in formula (6): ; (6).

[0023] The advantages and beneficial effects of this invention are as follows: 1. This invention is designed with modularity and simplicity as its core principles. The method does not rely on complex system modeling; instead, it rapidly constructs a basic model using a small amount of historical operational data. It directly expresses the water supply temperature as a function of load and flow rate. Unlike traditional empirical curves, this method extracts key thermal features reflecting the inertia of the heating network and user behavior from historical operational information and embeds these features as core model parameters into the water supply temperature setting process. In this way, determining the water supply temperature is not only intuitive and easy to implement but also accurately reflects the matching law between load and flow rate in the actual operating environment, effectively avoiding energy waste caused by overheating. This method requires few input parameters and has a clear structure, greatly reducing the implementation threshold. It facilitates rapid deployment in different types of heating systems and provides a reliable foundation for subsequent zoned, time-based, and temperature-based on-demand heating regulation.

[0024] 2. This invention offers flexible usage options to meet different needs. Each module is independent, and the modules can be combined according to actual requirements. For the new heating season, it has the following application scenarios: Mode 1: Model application phase ①, input the new weekly or monthly planned heat load and the upper limit of flow allowed by the system into the model, and generate the water supply temperature set curve for the whole cycle through real-time correction and optimization by the model, which is used to control the water temperature fluctuation trend in the long-term planning phase.

[0025] Mode 2: Based on Mode 1, the model application stage ② is carried out, which involves collecting a small amount of real-time data from the past few days or weeks to refine the set curve and improve the accuracy of short-term prediction and tracking.

[0026] Mode 3: When significant changes in operating conditions occur during operation (such as heat source switching, pipeline topology adjustment, sudden changes in load on the heat user side, etc.), the model is quickly reconstructed according to steps S1–S6, and either Mode 1 or Mode 2 is selected for implementation based on the actual operating status.

[0027] 3. This invention possesses dynamic correction and adaptive capabilities, constructing an intelligent closed-loop control system: It innovatively introduces a thermal feature recognition and correction mechanism, enabling self-learning and dynamic adjustment through real-time feedback of system operating data, thus achieving online updates of model parameters. This feature endows the system with proactive disturbance rejection, self-adaptation, and self-optimization capabilities, maintaining prediction and control accuracy even under uncertain operating conditions such as load fluctuations, user behavior differences, or environmental disturbances.

[0028] 4. Small data requirements for modeling, reducing reliance on real-time data: This method only requires historical data from the previous heating season (such as flow rate, load, and water supply temperature) to complete the basic modeling, and uses a small amount of real-time data at the beginning of the new season to quickly update and calibrate the model. This avoids reliance on high-frequency real-time data throughout the year, significantly reducing data acquisition and processing costs, while also allowing for rapid adaptation to the current heating season conditions and improving the timeliness of predictive response.

[0029] 5. Adapt to production plan values ​​and strengthen system energy consumption benchmarking control: Unlike the existing traditional heating control strategy that mainly relies on meteorological compensation, has a delayed response, and is extensive in adjustment, this method can combine system operation plans (such as daily planned load, scheduling schemes, etc.) to adjust the water supply temperature in real time through intelligent algorithms while ensuring the thermal comfort of users, so as to achieve precise matching between the heating capacity of the heat source and the heat load demand of users.

[0030] The intelligent water supply temperature control model constructed in this invention has predictive, adaptive and iterative correction capabilities, and can flexibly cope with uncertainties such as load fluctuations, differences in user behavior and environmental disturbances, significantly improving the control accuracy and system response efficiency, and providing a more intelligent operation optimization method for centralized heating systems.

[0031] 6. Simple model structure, few control variables, and low implementation cost: The water supply temperature model of this invention has a simple structure, involving only three commonly obtained physical parameters in the heating system: heating load, flow rate, and water supply temperature. It does not require the introduction of additional sensors or complex measurement methods, nor does it involve civil engineering modifications to existing heating networks or heat source systems. It possesses extremely high feasibility and engineering feasibility, making it easy to promote and apply in various heating systems.

[0032] 7. Solving the problem of limited applicable scenarios and having good model transfer and generalization capabilities: This method focuses on the three most basic and universally available parameters in heating systems: water supply temperature, flow rate and heat load. It does not rely on complex or high-frequency data acquisition at all. It can be deployed and used regardless of whether the system has completed digital transformation, and has strong versatility, compatibility and engineering scalability.

[0033] 8. Short deployment cycle and convenient system integration: This method does not rely on specialized software and hardware platforms and can be integrated and deployed on existing SCADA or EMS systems. Data acquisition, model operation, and temperature setting can be quickly connected through conventional industrial control interfaces. The modeling and control process is clear, the deployment cycle is short, and it is easy to quickly put into use in engineering practice.

[0034] 9. Integrating intelligent control with traditional physical mechanisms to improve the level of intelligent regulation: This method takes into account both data-driven and thermal principles, constructs prediction logic with physical boundaries as constraints, and builds a dual-channel regulation architecture of "intelligent prediction + intelligent correction" through multi-model integration and residual feedback, which is adapted to the trend of centralized heating systems transforming to intelligent scheduling. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the control system of the present invention; Figure 2 This is a flowchart illustrating the control process of the present invention; Figure 3 This is an application flowchart of an embodiment of the present invention; Figure 4 This is a graph showing the changes in simulated water temperature and actual simulated water temperature for different models during the heating season 2 of this invention. Figure 5 This is a graph showing the changes in simulated water temperature and actual water temperature after real-time correction and optimization during the heating season 2, as described in this invention. Figure 6 This is a bar chart showing the target and actual (14 days) water supply temperature values ​​in an embodiment of the present invention. Detailed Implementation

[0036] The present invention will be further described in detail below through specific embodiments. The following embodiments are merely descriptive and not limiting, and should not be used to limit the scope of protection of the present invention.

[0037] An innovative method for intelligent water supply temperature control based on thermal feature analysis and migration correction addresses the dynamic matching problem between user demand and operational capacity in heating systems. Unlike traditional methods relying on experience or fixed curves, this method emphasizes using predicted load or production plan values ​​as core inputs, combined with real-time system flow constraints, to directly generate a water supply temperature setpoint that meets demand. Based on thermal engineering mechanisms and modular modeling, the method constructs a lightweight control model, reducing water supply temperature to a function of heat load and flow rate. This avoids complex state variable calculations while maintaining strong adaptability and versatility.

[0038] The aforementioned control method combines fundamental principles of thermodynamics with a thermal characteristic recognition mechanism to establish a simplified yet efficient water supply temperature prediction and control model, enabling rapid modeling and regulation based on two core quantities: load and flow rate. Figure 1 As shown, the control system employed in the control method includes a peripheral module, a control module, and a controlled object. The peripheral module provides historical / real-time data input, the control module completes model construction and calibration, model application, and outputs control commands. The controlled object responds to the commands, executes operational adjustments, and generates operational demand feedback. Data acquisition in the peripheral module relies on flow sensors, heat meters, and water supply temperature sensors deployed at key nodes of the heating network. The data acquisition module obtains real-time operating data of the heating system, including water supply temperature. t g (°C), heat load Q (MW·h) and flow rate G (m) 3 The collected data is transmitted to the host computer and stored in the database, serving as the basis for modeling and prediction; like Figure 2 As shown, the steps of the control method are as follows: S1. Feature selection based on thermal engineering principles: Starting from the basic heat balance principle of the heating system, we search for more universal input features and determine the input features with the goal of parameter simplification. S2. Data Acquisition and Processing: By reading the database in the external modules and based on the actual historical data of the heating network operation, a parameter set with a consistent time scale is formed, including water supply temperature. t g ,flow G and heat load Q ; S3. Determine the basic model: Use parameter set data for basic model parameter calibration, and combine multiple regression or ensemble learning methods to establish specific functional relationships between water supply temperature, heat load, and flow rate, thus establishing the basic model. f 0: T_static=f 0( Q , G ); S4. Thermal Feature Identification: Input the parameter set into the basic model to obtain the simulated water temperature. T_static With respect to the actual water supply temperature t g The deviation is defined as the residual term characterizing the time-varying drift of the system's thermal characteristics, and a residual mapping model is established. ε = f 1( Q , G Furthermore, a multi-model fusion learning framework is introduced to further improve learning efficiency. f 1. Conduct training and integration to achieve adaptive representation of uncertain disturbances and drift characteristics under working conditions, and construct a thermal feature recognition module; S5. Model Combination: Combining the basic model f 0 and thermal feature recognition module f 1. Combining these elements, we obtain a water supply temperature prediction and control model, the output of which is: T_pred = T_static + ε = f 0( Q , G )+ f 1( Q , G ); S6. Model Application: Based on the actual situation during the heating season, it is divided into two stages: Phase 1 – Initial commissioning phase of the new heating season: Based on the planned heat load and flow limit of the hydraulic regulation of the heating system set in the peripheral modules, the water supply temperature prediction and control model will output the water supply temperature setpoint of the control unit. T_set ; Phase ② – Entering the new heating season operation phase: Collect actual operating data from the past few days (more than 2 days) or several weeks, and adjust the setpoints based on actual load and flow. T_set Perform verification and reset ， By incorporating actual water supply temperature, a rolling window retraining and automatic parameter update strategy is adopted to identify and incorporate new thermal features, and a residual correction function is constructed. e = f 2 ( Q , G ), and use it for T_set Perform real-time correction and optimization to form T_set = f 0( Q , G )+ f 1( Q , G )+ f 2( Q , G ); in, f 2 Characterization runtime relative to f The incremental correction term of 1 is used to compensate for additional drift and disturbance under the new operating conditions.

[0039] Moreover, S1 specifically refers to: Based on the fundamental heat balance principle of a heating system, under ideal and stable operating conditions (ignoring heat loss during the heat network transmission process), the heat output from the heat source should perfectly match the actual heat load demand at the user end. The heat conservation relationship is as follows: (1) (2) (3) (4) In the formula, Q 1. Q 2 and Q 3 represents the actual heat load of the building, the heat dissipation of the heat dissipation equipment, and the load on the energy station side, respectively; q The building heat load index is expressed in W / (m²). 3 ·℃); V is the external structural volume of the building, m 3 ; t in and t out Indoor and outdoor temperatures, respectively, in °C; K It is the building heat transfer coefficient, W / (m²). 2 ·℃); A Indicates the building's heating area, in meters. 2 ; c The specific heat capacity of hot water is J / (kg·℃). G The energy station's flow rate is kg / h. t g and t h These are the supply water temperature and the return water temperature, respectively, in °C; t p The average temperature of the supply and return water is expressed by the following formula: t p = ( t g + t h ) / 2; When the system is identified as a heating system, formulas (1) to (4) are simplified to: (5) To further explore the correlation between water supply temperature and other parameters and to find more general input characteristics, the Pearson correlation coefficient was used to analyze the correlation between water supply temperature and the parameters described in formula (5). With the goal of simplification, for return water temperature, the correlation coefficient between the supply and return water temperature difference and the load reached a high level, indicating that the load can also reflect the change in the supply and return water temperature difference. The water supply temperature is lightly described by the relationship presented in formula (6): ; (6).

[0040] To enhance the model's dynamic adaptability, this invention introduces a thermal feature identification and migration correction mechanism: by analyzing historical data to extract key features such as system thermal inertia, user heating patterns, and pipeline characteristics, and continuously integrating newly acquired data during operation, the model is corrected and iteratively optimized online. This mechanism enables water supply temperature control to proactively respond to load disturbances, differences in user behavior, and changes in environmental conditions, effectively offsetting and compensating for uncertainties.

[0041] At the engineering application level, this method has extremely low hardware requirements, relying only on common heat meters, flow meters, and temperature sensors to complete data acquisition. Through database access and modular calculations, it quickly generates water supply temperature setpoints and can be flexibly integrated with automatic controllers or operation and maintenance auxiliary decision-making platforms. Compared to traditional methods, this invention has the advantages of simple structure, quick deployment, and low cost, making it particularly suitable for the rapid integration and promotion of multi-level, multi-scenario heating systems. It supports a "heating on demand" adjustment strategy based on time, zone, and temperature, which not only optimizes heating quality but also effectively reduces energy consumption and pipeline losses, significantly improving the overall system operating efficiency.

[0042] This method is independent of specific heat source types and is applicable to multi-level quality regulation and control scenarios such as energy centers, heat exchange stations, and terminal buildings. It can be flexibly deployed in systems that are already digitized or not. Its core lies in constructing a simple, computationally lightweight, and data-dependent intelligent control model through thermal mechanism modeling and data-driven feature recognition. This model possesses self-learning, self-adaptive, and self-optimizing capabilities, enabling dynamic response and efficient matching to changes in operating load and flow, significantly improving supply and demand coordination and system stability. By integrating historical features with real-time data, it achieves a shift from traditional experience-based scheduling to intelligent prediction and adaptive control, demonstrating good system compatibility and broad application potential.

[0043] This invention's embodiments are based on actual operational data of a centralized heating system, using an energy station in Tianjin as the research object. This heating system consists of a primary network directly powered by boilers, with a total heating area exceeding 240,000 square meters. The heating needs include various facilities such as office buildings, research buildings, dormitories, canteens, and laboratories. This embodiment primarily uses actual daily data from two consecutive heating seasons (described as Heating Season 1 and Heating Season 2) of this energy station as data support. The specific process is as follows... Figure 3 As shown in the figure. This embodiment specifically uses various machine learning models for training and prediction. The specific implementation process includes the following steps: (a) Data preprocessing: Read the database and form a parameter set with consistent time scale based on the actual historical data of the heating network operation.

[0044] (II) Establishment of basic model: In this embodiment, the polynomial regression method is used to establish a basic model to make preliminary predictions on water supply temperature and derive the formula. Specifically, data from heating season 1 are selected to calibrate the parameters of the basic model.

[0045] (III) Thermal Feature Recognition: A thermal feature recognition module is established using historical data. Specifically, data from the first heating season is selected, and the water supply temperature calculated by the basic model is compared with the actual water temperature. Machine learning algorithms such as Random Forest (RF), Gradient Boosting Decision Tree (XGBoost), and Multilayer Perceptron (MLP) are used for Stacking algorithm fusion training to construct the thermal feature recognition module. (iv) Model combination: The basic model is combined with the thermal feature recognition module to obtain the water supply temperature prediction and control model.

[0046] (V) Model Application: Stage ① uses the data from heating season 2 as the application set to evaluate the model's predictive ability and stability in future periods. In this example, the data from heating season 2 is only used as application data and is not involved in the calibration of model parameters; Stage ② In this embodiment, the actual water supply temperature data is used to perform real-time correction and optimization of the model's set curve, and based on the planned load and system flow constraints, the target water supply temperature values ​​for the corresponding target load in the next two weeks are obtained.

[0047] To comprehensively evaluate the predictive performance of the constructed model and better demonstrate its effectiveness, a comparative analysis was conducted on the application set using only the basic model and the combined thermal feature recognition module (water supply temperature prediction and control model). This was achieved by comparing the prediction curves and error indices (MAE, R²) of the two models on the application set. 2 The effect can more intuitively demonstrate the advantages of the multi-model residual correction strategy in the dynamic prediction task of heating system. Figure 4The graph shows a comparison between the actual water supply temperature and the simulation results of the two models during the application period. As can be seen from the graph, the basic model can roughly follow the temperature change trend, but it has a certain lag and deviation when the system load fluctuates greatly or the temperature changes abruptly. However, the combined thermal feature identification module (water supply temperature prediction and control model) shows a stronger trend fitting ability throughout the entire prediction interval, and can more accurately capture temperature fluctuations. The simulation results are in high agreement with the measured values.

[0048] Furthermore, Table 1 presents the simulation curves and error index simulation results for different models, which can further verify the improvement in accuracy of the models. The MAE of the basic model on the application set is 1.394℃, and the coefficient of determination R0 is... 2 The MAE of 0.683 indicates that it can characterize the temperature trend to some extent, but the error is relatively large. However, after combining the thermal feature recognition module, the MAE of the model decreased to 0.633℃, a reduction of over 54.6%, while R... 2 The goodness of fit was significantly improved to 0.933.

[0049] Table 1 Simulation performance of different models

[0050] Validation during the second heating season showed that the proposed method exhibits good generalization ability and prediction stability under actual operating conditions. Compared with the simple steady-state model, the predicted MAE for water supply temperature decreased from 1.394℃ to 0.633℃, and R... 2 The accuracy of the prediction was significantly improved, increasing from 0.683 to 0.931. The model relies solely on standard operating parameters, requiring no additional sensors or high-frequency data input, demonstrating good engineering feasibility and applicability, making it suitable for the rapid retrofitting and upgrading of existing heating systems. Future research can incorporate outdoor meteorological factors, building thermal inertia characteristics, and indoor temperature feedback mechanisms to further enhance the model's overall adaptability and control effectiveness, providing technical support for promoting energy conservation, carbon reduction, and intelligent development of centralized heating systems.

[0051] To examine the actual predictive guidance effect, further real-time correction and optimization were performed. For heating season 2, actual water supply temperature data was input into the model for real-time correction and optimization. After determining the optimal rolling period to be 3 days, the prediction accuracy was significantly improved after real-time correction and optimization, with MAE decreasing to 0.302 ℃ and R... 2 The error was improved to 0.987, a reduction of approximately 58.8% compared to the previous value. This indicates that real-time correction can effectively compensate for time-varying deviations and disturbances during system operation, resulting in stronger consistency in the model's tracking of actual operating conditions and more stable and reliable prediction results. (See Table 2 and...) Figure 5 As shown.

[0052] Meanwhile, to simulate the water supply temperature prediction and setting process under future engineering application scenarios, the model was first verified and corrected using one week of actual operating data. Then, combining the planned heat load provided by the external modules and the flow constraints of the system's hydraulic regulation, the predicted water supply temperature for the next two weeks under the corresponding planned load conditions was calculated, and the target water supply temperature setting was formed accordingly. For example... Figure 6 As shown, it can effectively provide quantitative guidance for water supply temperature setting based on planned load and flow constraints.

[0053] Table 2 Real-time Correction and Optimization Simulation Performance

[0054] Although embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A method for intelligent control of water supply temperature based on thermal feature analysis and migration correction, characterized in that: The aforementioned control method combines fundamental principles of thermal engineering with a thermal characteristic recognition mechanism to establish a simplified yet efficient water supply temperature prediction and control model, enabling rapid modeling and regulation based on two core quantities: load and flow rate. The control system employed in this method includes a peripheral module, a control module, and a controlled object. The peripheral module provides historical / real-time data input, while the control module completes model construction and calibration, model application, and outputs control commands. The controlled object responds to commands, executes operational adjustments, and generates operational demand feedback. Data acquisition in the peripheral module relies on flow sensors, heat meters, and water supply temperature sensors deployed at key nodes of the heating network. The data acquisition module obtains real-time operating data of the heating system, including water supply temperature. t g Heat load Q and traffic G The collected data is transmitted to a host computer and stored in a database, serving as the basis for modeling and prediction. The steps of the control method are as follows: S1. Feature selection based on thermal engineering principles: Starting from the basic heat balance principle of the heating system, we search for more universal input features and determine the input features with the goal of parameter simplification. S2. Data Acquisition and Processing: By reading the database in the external modules and based on the actual historical data of the heating network operation, a parameter set with a consistent time scale is formed, including water supply temperature. t g ,flow G and heat load Q ; S3. Determine the basic model: Use parameter set data for basic model parameter calibration. Combine multiple regression or ensemble learning methods to establish the functional relationship between water supply temperature, heat load, and flow rate, and determine the basic model. f 0: T_static=f 0( Q , G )； S4. Thermal Feature Identification: Input the parameter set into the basic model to obtain the simulated water temperature. T_static With respect to the actual water supply temperature t g The deviation is defined as the residual term characterizing the time-varying drift of the system's thermal characteristics, and a residual mapping model is established. f 1: ε = f 1( Q , G )； Introducing a multi-model fusion learning framework f 1. Conduct training and integration to achieve adaptive representation of uncertain disturbances and drift characteristics under working conditions, and construct a thermal feature recognition module; S5. Model Combination: Combining the basic model f 0 and residual mapping model f 1. Combining these elements, we obtain a water supply temperature prediction and control model: T_pred = T_static + ε = f 0( Q , G )+ f 1( Q , G )； S6. Model Application: Divided into two stages based on the actual situation during the heating season: Phase 1 – Initial commissioning phase of the new heating season: Based on the planned heat load and flow limit of the hydraulic regulation of the heating system set in the peripheral modules, the water supply temperature prediction and control model will output the water supply temperature setpoint of the control unit. T_set ; Phase ② – Entering the new heating season operation phase: Collect actual operating data from the past few days (more than 2 days) or several weeks, and adjust the setpoints based on actual load and flow. T_set Perform verification and reset ， By incorporating actual water supply temperature, a rolling window retraining and automatic parameter update strategy is adopted to identify and incorporate new thermal features, and a residual correction function is constructed. e = f 2 ( Q , G ), and use it for T_set Perform real-time correction and optimization to form T_set = f 0( Q , G )+ f 1( Q , G )+ f 2 ( Q , G )； in, f 2 Characterization runtime relative to f The incremental correction term of 1 is used to compensate for additional drift and disturbance under the new operating conditions.

2. The intelligent water supply temperature control method based on thermal feature analysis and migration correction according to claim 1, characterized in that: Specifically, S1 is: Based on the fundamental heat balance principle of a heating system, under ideal and stable operating conditions (ignoring heat loss during the heat network transmission process), the heat output from the heat source should perfectly match the actual heat load demand at the user end. The heat conservation relationship is as follows: ； (1) ； (2) ； (3) ； (4) In the formula, Q 1. Q 2 and Q 3 represents the actual heat load of the building, the heat dissipation of the heat dissipation equipment, and the load on the energy station side, respectively; q The building heat load index is expressed in W / (m²). 3 ·℃); V is the external structural volume of the building, m 3 ; t in and t out Indoor and outdoor temperatures, respectively, in °C; K It is the heat transfer coefficient of the heat dissipation device, W / (m²). 2 ·℃); A The heat dissipation area of ​​the heat dissipation device is expressed in meters (m). 2 ; c The specific heat capacity of hot water is J / (kg·℃). G The energy station's flow rate is kg / h. t g and t h These are the supply water temperature and the return water temperature, respectively, in °C; t p The average temperature of the supply and return water is expressed by the following formula: t p = ( t g + t h ) / 2; When the system is identified as a heating system, formulas (1) to (4) are simplified to: ； (5) To further explore the correlation between water supply temperature and other parameters and to find more general input characteristics, the Pearson correlation coefficient was used to analyze the correlation between water supply temperature and the parameters described in formula (5). With the goal of simplification, for return water temperature, the correlation coefficient between the supply and return water temperature difference and the load reached a high level, indicating that the load can also reflect the change in the supply and return water temperature difference. The water supply temperature is lightly described by the relationship presented in formula (6): ； (6)。

Images