Low-vacuum circulating water heating dynamic load matching method based on AI regulation and control

Abstract

The invention discloses a low-vacuum circulating water heating dynamic load matching method based on AI regulation and control, and the method comprises the following steps: step S100, multi-source data real-time collection: building thermal environment parameters are synchronously collected through a temperature sensor array, a vacuum pressure transmitter, a flowmeter and an environment monitoring unit which are deployed in a distributed manner; according to the method, the control period is compressed to the minute level, compared with traditional PID control, the response speed is increased by several times, the indoor temperature can be controlled more accurately, and meanwhile the failure rate of equipment is reduced. According to the sudden weather coping mechanism, heat supply strengthening can be started 30 minutes ahead of time under extreme weather such as snowstorm weather, and the stability of the system is ensured. The method is characterized in that the prediction advantage of deep learning and the optimization capability of model prediction control are deeply combined, and the control problem of multivariable strong coupling of the low vacuum system is solved through a verification system of multi-source data fusion and virtual-real combination.

Description

Technical Field

[0001] The present application relates to the field of heating technology, and specifically to a dynamic load matching method for low-vacuum circulating water heating based on AI control. Background Art

[0002] In the heating sector, traditional low-vacuum heating systems once held a significant market share. Their control strategies often employed fixed-load regulation or simple temperature feedback mechanisms. However, with the increasing complexity of building heating needs and rising demands for heating quality and energy efficiency, the drawbacks of these traditional technologies have become increasingly apparent.

[0003] On the one hand, traditional systems are unable to cope with the dynamic changes in building heat load. The heat load of a building is not static. It is affected by multiple factors such as outdoor meteorological conditions (such as temperature, wind speed, solar radiation, etc.), building usage (personnel flow, equipment operation, etc.), and changes in the performance of the building envelope. However, the existing system lacks an effective prediction mechanism and cannot perceive the heat load fluctuations caused by these factors in advance, which inevitably leads to problems of delayed or excessive heating. Delayed heating will prevent the indoor temperature from reaching the comfort standard for a period of time, affecting the user's living experience; while excessive heating will cause energy waste, increase heating costs, and may also have an adverse effect on indoor air quality and equipment life.

[0004] On the other hand, conventional PID control performs poorly in complex and changing operating conditions. PID control is a classic control algorithm, but when faced with the complex operating conditions of traditional low-vacuum heating systems, characterized by nonlinearity, time-varying behavior, and significant hysteresis, its parameters struggle to accurately adjust in real time to accommodate system changes. This results in the system operating in a suboptimal state for extended periods, resulting in low heating efficiency and low energy utilization, making it difficult to meet the demands of modern heating systems for efficient, energy-saving, and stable operation. Summary of the Invention

[0005] To this end, the present application provides a low-vacuum circulating water heating dynamic load matching method based on AI control to solve the problems of low load prediction accuracy and insufficient energy efficiency optimization in the existing technology.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] In a first aspect, a dynamic load matching method for low-vacuum circulating water heating based on AI control comprises the following steps:

[0008] Step S100, real-time multi-source data collection: using a distributed array of temperature sensors, vacuum pressure transmitters, flow meters, and environmental monitoring units, synchronously collect building thermal environment parameters, equipment status parameters, and external meteorological parameters;

[0009] Step S200, dynamic load forecasting: the collected real-time data is integrated with the historical operation database for spatiotemporal features, and input into the pre-trained AI neural network model to generate an hourly heat load forecast curve for the next 2 hours.

[0010] Step S300, multi-parameter coordinated control: Based on the model predictive control (MPC) framework, an optimal control instruction set is generated according to the predicted load curve, including:

[0011] a) Adjust the operating state of the vacuum maintaining device through the electric throttle valve to maintain the system vacuum degree in the range of 10-30kPa;

[0012] b) Control the output frequency of the circulating pump inverter to match the circulating flow rate with the current heat load demand;

[0013] c) Adjust the opening combination of heat exchanger valves to optimize the heat distribution ratio of each heat exchange unit;

[0014] Step S400, closed-loop optimization and learning: collect actual operating data to build a dynamic knowledge base, and perform an online learning process every 24 hours, including calculating the real-time deviation value of the system energy efficiency ratio (COP), analyzing equipment health indicators and generating maintenance recommendation reports, and using incremental learning algorithms to update the weight parameters of the neural network model.

[0015] Preferably, the temperature sensor array comprises:

[0016] The first sensor cluster deployed on the building's exterior wall uses high-temperature resistant optical fiber temperature measurement units;

[0017] The second sensor cluster is set up in the core area of the room, including at least 20 evenly distributed wireless temperature nodes;

[0018] A third sensor cluster installed on the outer wall of the heating pipe is equipped with a contact thermocouple array;

[0019] The environmental monitoring unit obtains the weather forecast for the next 6 hours through a weather data receiving device and uses an infrared thermal imager to monitor heat loss at the building entrance and exit.

[0020] Preferably, the AI neural network model adopts an LSTM-Transformer hybrid architecture and performs the following processing:

[0021] Analyze the periodic variation of heating load through the time feature extraction layer;

[0022] The mapping relationship between vacuum degree fluctuation and indoor temperature distribution is established through the spatial correlation analysis layer;

[0023] The output includes the prediction results with confidence intervals and the weight distribution of key influencing factors.

[0024] Preferably, the training process of the AI neural network model in step S200 includes:

[0025] Construct a historical dataset with seasonal characteristic annotations, with a time span of no less than 5 heating seasons;

[0026] The sliding window method is used to generate training samples, with a window length of 72 hours and including a complete day and night cycle;

[0027] Design a dual attention mechanism to handle meteorological parameter mutation events and building thermal inertia characteristics respectively;

[0028] Reuse the underlying parameters of the pre-trained building thermal response model through transfer learning.

[0029] Preferably, step S300 specifically includes three control modes:

[0030] Mode I: When the predicted load fluctuation is less than 10%, PID parameter self-tuning control based on fuzzy rules is adopted;

[0031] Mode II: When a heavy snow weather warning is detected, the feedforward-feedback composite control strategy is activated to increase the heating power 30 minutes in advance;

[0032] Mode III: When the vacuum degree is lower than 15kPa for 10 minutes, the system switches to the standby vacuum pump group and issues an equipment maintenance alarm.

[0033] Preferably, the method further includes a fault diagnosis step, wherein the fault diagnosis step is used to extract the current harmonic characteristics of the circulating pump, the valve action response curve and the temperature gradient data of the heat exchanger in real time;

[0034] A random forest classifier is used to identify early failure modes. When a failure occurs, an early warning mechanism is triggered. The early warning mechanism includes:

[0035] Level 1 warning: When the energy efficiency deviation value exceeds 5% for three consecutive cycles, a parameter calibration prompt is issued;

[0036] Level 2 warning: When mechanical wear characteristics are detected, the equipment load rate is automatically reduced by 20%;

[0037] Level 3 warning: When a vacuum leak risk is diagnosed, the system will be shut down for emergency protection.

[0038] Preferably, the method further includes a user interaction step;

[0039] Provide a visual interface to display the three-dimensional thermal map of heating intensity in each area;

[0040] Receive temperature preference parameters set by the user and balance comfort and energy consumption through a multi-objective optimization algorithm;

[0041] Generate daily energy efficiency analysis reports, marking the equipment and time periods with the greatest energy saving potential.

[0042] Preferably, the method further includes a regional coordination control step:

[0043] Divide the building into multiple independent control zones, and set personalized temperature control strategies for each zone;

[0044] Dynamically adjust the flow distribution of each branch through the hydraulic balance algorithm to eliminate uneven hot and cold pipe network;

[0045] When it is detected that a certain area has a long period of low load, the corresponding branch will be automatically closed and anti-freeze protection will be activated.

[0046] Preferably, the method further includes an energy efficiency optimization step:

[0047] Store excess heat during off-peak hours in the grid, and dynamically adjust the amount of heat storage based on weather forecasts;

[0048] Implement an intelligent rotation strategy for circulating pumps to balance the cumulative operating time of each pump within 10%;

[0049] A pressure fluctuation suppression algorithm is used to control the pressure difference fluctuation at key nodes of the pipeline network within the range of ±5%.

[0050] Compared with the prior art, this application has at least the following beneficial effects:

[0051] This method compresses the control cycle to minutes, significantly increasing the response speed several times compared to traditional PID control. It enables more precise control of indoor temperature while reducing equipment failure rates. The sudden weather response mechanism can initiate enhanced heating 30 minutes in advance during extreme weather conditions such as blizzards, ensuring system stability. The innovation of this method lies in its deep integration of the predictive advantages of deep learning with the optimization capabilities of model predictive control. Through multi-source data fusion and a virtual-real verification system, it addresses the control challenges of tightly coupled multivariable systems in low-vacuum systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] To more intuitively illustrate the prior art and the present application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be considered as limiting conditions for implementing the present application; for example, based on the technical concepts disclosed in this application and the exemplary drawings, those skilled in the art are capable of easily making routine adjustments or further optimizations to the addition / reduction / attribution division, specific shapes, positional relationships, connection methods, dimensional ratios, etc. of certain units (components).

[0053] Figure 1A flow chart of a dynamic load matching method for low-vacuum circulating water heating based on AI control provided in this application. DETAILED DESCRIPTION

[0054] The present application will be further described below in detail through specific embodiments in conjunction with the accompanying drawings.

[0055] A dynamic load matching method for low-vacuum circulating water heating based on AI control includes the following steps:

[0056] Step S100, real-time multi-source data collection: using a distributed array of temperature sensors, vacuum pressure transmitters, flow meters, and environmental monitoring units, synchronously collect building thermal environment parameters, equipment status parameters, and external meteorological parameters;

[0057] The temperature sensor array comprises:

[0058] The first sensor cluster deployed on the building's exterior wall uses high-temperature resistant optical fiber temperature measurement units;

[0059] The second sensor cluster is set up in the core area of the room, including at least 20 evenly distributed wireless temperature nodes;

[0060] A third sensor cluster installed on the outer wall of the heating pipe is equipped with a contact thermocouple array;

[0061] The environmental monitoring unit obtains the weather forecast for the next 6 hours through a weather data receiving device and uses an infrared thermal imager to monitor heat loss at the building entrances and exits;

[0062] Step S200, dynamic load forecasting: the collected real-time data is integrated with the historical operation database for spatiotemporal features, and input into the pre-trained AI neural network model to generate an hourly heat load forecast curve for the next 2 hours.

[0063] The AI neural network model uses an LSTM-Transformer hybrid architecture and performs the following processing:

[0064] Analyze the periodic variation of heating load through the time feature extraction layer;

[0065] The mapping relationship between vacuum degree fluctuation and indoor temperature distribution is established through the spatial correlation analysis layer;

[0066] Output includes the prediction results of confidence intervals and the weight distribution of key influencing factors;

[0067] Step S300, multi-parameter coordinated control: Based on the model predictive control (MPC) framework, an optimal control instruction set is generated according to the predicted load curve, including:

[0068] a) Adjust the operating state of the vacuum maintaining device through the electric throttle valve to maintain the system vacuum degree in the range of 10-30kPa;

[0069] b) Control the output frequency of the circulating pump inverter to match the circulating flow rate with the current heat load demand;

[0070] c) Adjust the opening combination of heat exchanger valves to optimize the heat distribution ratio of each heat exchange unit;

[0071] Step S400, closed-loop optimization and learning: collect actual operating data to build a dynamic knowledge base, and perform an online learning process every 24 hours, including calculating the real-time deviation value of the system energy efficiency ratio (COP), analyzing equipment health indicators and generating maintenance recommendation reports, and using incremental learning algorithms to update the weight parameters of the neural network model.

[0072] This method achieves precise dynamic load matching for low-vacuum heating systems by constructing a fully closed-loop intelligent control system combining "perception-prediction-control-optimization." First, during the data collection phase, a three-tiered temperature sensor cluster (exterior wall surface, interior core area, and pipeline exterior) is innovatively employed in conjunction with an environmental monitoring unit. Combined with infrared thermal imaging technology, this method achieves three-dimensional perception of the building's thermal environment (improving the spatial resolution of temperature monitoring).

[0073] In the load forecasting stage, an AI neural network model is used to capture the diurnal rhythm of the heating load through the time feature extraction layer, and the spatial correlation analysis layer is used to establish a dynamic mapping of vacuum fluctuations and temperature distribution, greatly improving the accuracy of load forecasts for the next 2 hours.

[0074] The model predictive control (MPC) framework is introduced in the multi-parameter coordinated regulation stage, combined with digital twin verification technology to simultaneously optimize the matching relationship between vacuum degree (10-30kPa), circulation flow (2-8m3 / h) and heat exchange power (50-200kW), thereby significantly improving the system energy efficiency ratio (COP).

[0075] The closed-loop optimization mechanism performs incremental learning every 24 hours, dynamically adjusts the MPC cost function weights through reinforcement learning, and implements preventive maintenance in combination with equipment health assessment.

[0076] This method compresses the control cycle to minutes, significantly increasing the response speed several times compared to traditional PID control. It enables more precise control of indoor temperature while reducing equipment failure rates. The sudden weather response mechanism can initiate enhanced heating 30 minutes in advance during extreme weather conditions such as blizzards, ensuring system stability. The innovation of this method lies in its deep integration of the predictive advantages of deep learning with the optimization capabilities of model predictive control. Through multi-source data fusion and a virtual-real verification system, it addresses the control challenges of tightly coupled multivariable systems in low-vacuum systems.

[0077] The training process of the AI neural network model in step S200 includes:

[0078] Construct a historical dataset with seasonal characteristics annotations, covering no less than five heating seasons. Construct an annotated dataset covering five heating seasons, covering typical operating conditions such as severe cold and snowmelt periods, to improve the model's ability to identify different seasonal characteristics.

[0079] The sliding window method is used to generate training samples. The window length is 72 hours and includes a complete day and night cycle. A 72-hour sliding window (including 3 complete day and night cycles) is used to generate training samples to effectively capture the lag effect of building thermal inertia.

[0080] A dual attention mechanism is designed to handle meteorological parameter mutation events and building thermal inertia characteristics respectively. The dual attention mechanism handles meteorological mutations (such as cold waves) and building heat storage characteristics respectively, reducing the error of the prediction model in scenarios with sudden temperature drops;

[0081] By reusing the underlying parameters of the pre-trained building thermal response model through transfer learning, the application of transfer learning technology can greatly shorten the model training time while maintaining a high prediction accuracy.

[0082] Step S300 specifically includes three control modes:

[0083] Mode I: When the predicted load fluctuation is less than 10%, PID parameter self-tuning control based on fuzzy rules is adopted;

[0084] Mode II: When a heavy snow weather warning is detected, the feedforward-feedback composite control strategy is activated to increase the heating power 30 minutes in advance;

[0085] Mode III: When the vacuum degree is lower than 15kPa for 10 minutes, the system switches to the standby vacuum pump group and issues an equipment maintenance alarm.

[0086] The three-mode control strategy of the above technical solution covers all operating conditions: Mode I uses fuzzy PID parameter self-tuning to maintain control stability within a ±10% load fluctuation range, avoiding equipment wear and tear caused by frequent adjustments. Mode II establishes a feedforward-feedback composite control for extreme weather. When a heavy snowfall warning is received, the vacuum pump power is increased and the circulation flow rate is increased in advance to ensure that the indoor temperature fluctuates no more than 1°C during a sudden drop in air temperature. Mode III establishes dual vacuum safety thresholds (15kPa warning and 10kPa emergency protection). When a vacuum leak is detected, the backup unit is switched and the damaged branch is closed within 10 seconds, significantly improving the response speed compared to traditional protection mechanisms. The three modes switch smoothly and avoid control command conflicts.

[0087] The system also includes a fault diagnosis step, which is used to extract the harmonic characteristics of the circulating pump current, the valve action response curve, and the heat exchanger temperature gradient data in real time. The analysis of the 5th / 7th harmonic content of the circulating pump current helps to identify electrical faults such as broken rotor bars. The monitoring of the valve opening-flow response curve can diagnose actuator sticking. The tracking of the heat exchanger inlet and outlet temperature gradient can promptly detect scaling or blockage problems.

[0088] A random forest classifier is used to identify early failure modes. When a failure occurs, an early warning mechanism is triggered. The early warning mechanism includes:

[0089] Level 1 warning triggers automatic parameter calibration: When the energy efficiency deviation value exceeds 5% for three consecutive cycles, a parameter calibration prompt is issued;

[0090] Secondary warning starts equipment load reduction protection: when mechanical wear characteristics are detected, the equipment load rate is automatically reduced by 20%;

[0091] Level 3 early warning safety interlock: When a vacuum leak risk is diagnosed, the system will be shut down for emergency protection.

[0092] It also includes user interaction steps;

[0093] Provides a visual interface to display the three-dimensional thermal map of heating intensity in each area. The three-dimensional thermal map displays the building heat distribution with a 0.5m×0.5m grid accuracy, helping operation and maintenance personnel quickly locate insulation weaknesses;

[0094] Receive the temperature preference parameters set by the user and balance comfort and energy consumption through a multi-objective optimization algorithm. The multi-objective optimization algorithm integrates the user-set temperature (such as 22±1℃) with real-time electricity prices, equipment load rate and other parameters to generate an optimal solution set for the user to choose;

[0095] Generate a daily energy efficiency analysis report, marking the equipment and time periods with the greatest energy-saving potential. The energy efficiency report can help users control the heating system. For example, it can indicate that a heat exchange station has 15% excess heating between 10:00 and 12:00 every day, and recommend adjusting the valve opening combination.

[0096] It also includes regional coordinated control steps:

[0097] Divide the building into multiple independent control zones, and set personalized temperature control strategies for each zone to meet the needs of different areas;

[0098] Dynamically adjust the flow distribution of each branch through the hydraulic balance algorithm to eliminate uneven hot and cold pipe network;

[0099] When it is detected that a certain area has a long period of low load, the corresponding branch will be automatically closed and anti-freeze protection will be activated.

[0100] It also includes energy efficiency optimization steps:

[0101] Excess heat is stored during off-peak hours of the grid. The amount of heat storage is dynamically adjusted based on the weather forecast. Off-peak heat storage is combined with the thermal inertia of the building (time constant 4-6 hours). Excess heat is stored during low-price periods (heat storage temperature is increased to 85°C) and released during peak hours, thereby reducing electricity costs.

[0102] Implement an intelligent rotation strategy for circulating pumps to balance the cumulative operating time of each pump within 10%;

[0103] A pressure fluctuation suppression algorithm is used to control the pressure difference fluctuation at key nodes of the pipeline network within the range of ±5%, thereby reducing energy loss caused by hydraulic imbalance.

[0104] The technical features of the above embodiments can be combined arbitrarily (as long as there is no contradiction in the combination of these technical features). In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described; these embodiments that are not explicitly written should also be considered to be within the scope of this specification.

Claims

1. A dynamic load matching method for low vacuum circulating water heating based on AI control, characterized in that: The following steps are involved: Step S100, real-time multi-source data collection: using a distributed array of temperature sensors, vacuum pressure transmitters, flow meters, and environmental monitoring units, synchronously collect building thermal environment parameters, equipment status parameters, and external meteorological parameters; Step S200, dynamic load forecasting: the collected real-time data is integrated with the historical operation database for spatiotemporal features, and input into the pre-trained AI neural network model to generate an hourly heat load forecast curve for the next 2 hours. Step S300, multi-parameter coordinated control: Based on the model predictive control framework, the optimal control instruction set is generated according to the predicted load curve, specifically including: a) Adjust the operating state of the vacuum maintaining device through the electric throttle valve to maintain the system vacuum degree in the range of 10-30kPa; b) Control the output frequency of the circulating pump inverter to match the circulating flow rate with the current heat load demand; c) Adjust the opening combination of heat exchanger valves to optimize the heat distribution ratio of each heat exchange unit; Step S400, closed-loop optimization and learning: collect actual operating data to build a dynamic knowledge base, and perform an online learning process every 24 hours. The online learning process includes calculating the real-time deviation value of the system energy efficiency ratio, analyzing equipment health indicators and generating maintenance recommendation reports, and using incremental learning algorithms to update the weight parameters of the neural network model.

2. The dynamic load matching method for low vacuum circulating water heating based on AI control according to claim 1 is characterized in that: The temperature sensor array comprises: The first sensor cluster is deployed on the building's exterior wall. The first sensor uses a high-temperature resistant optical fiber temperature measurement unit. A second sensor cluster is set up in the indoor core area, and the second sensor cluster includes at least 20 evenly distributed wireless temperature nodes; a third sensor cluster installed on an outer wall of the heating pipe, the third sensor cluster being configured with a contact thermocouple array; The environmental monitoring unit obtains the weather forecast for the next 6 hours through a weather data receiving device and uses an infrared thermal imager to monitor heat loss at the building entrance and exit.

3. The dynamic load matching method for low vacuum circulating water heating based on AI control according to claim 1 is characterized in that: The AI neural network model uses an LSTM-Transformer hybrid architecture and performs the following processing steps: Analyze the periodic variation of heating load through the time feature extraction layer; The mapping relationship between vacuum degree fluctuation and indoor temperature distribution is established through the spatial correlation analysis layer; The output includes the prediction results with confidence intervals and the weight distribution of key influencing factors.

4. The dynamic load matching method for low vacuum circulating water heating based on AI control according to claim 1 is characterized in that: The training process of the AI neural network model in step S200 includes: Construct a historical dataset with seasonal characteristic annotations, with a time span of no less than 5 heating seasons; The sliding window method is used to generate training samples, with a window length of 72 hours and including a complete day and night cycle; Design a dual attention mechanism to handle meteorological parameter mutation events and building thermal inertia characteristics respectively; Reuse the underlying parameters of the pre-trained building thermal response model through transfer learning.

5. The method for dynamic load matching of low vacuum circulating water heating based on AI control according to claim 1 is characterized in that: Step S300 specifically includes three control modes: Mode I: When the predicted load fluctuation is less than 10%, PID parameter self-tuning control based on fuzzy rules is adopted; Mode II: When a snowstorm warning is detected, the feedforward-feedback composite control strategy is activated to increase the heating power 30 minutes in advance; Mode III: When the vacuum degree is lower than 15kPa for 10 minutes, the system switches to the standby vacuum pump group and issues an equipment maintenance alarm.

6. The dynamic load matching method for low vacuum circulating water heating based on AI control according to claim 1 is characterized in that: The method further includes a fault diagnosis step for extracting the current harmonic characteristics of the circulating pump, the valve action response curve and the temperature gradient data of the heat exchanger in real time; A random forest classifier is used to identify early failure modes. When a failure occurs, an early warning mechanism is triggered. The early warning mechanism includes: Level 1 warning: When the energy efficiency deviation value exceeds 5% for three consecutive cycles, a parameter calibration prompt is issued; Level 2 warning: when mechanical wear characteristics are detected, the equipment load rate is automatically reduced by 20%; Level 3 warning: when vacuum leakage risk is diagnosed, the system will be shut down for emergency protection.

7. The method for dynamic load matching of low vacuum circulating water heating based on AI control according to claim 1 is characterized in that: It also includes user interaction steps; The user interaction steps provide a visual interface for displaying a three-dimensional thermal map of the heating intensity of each area; receive the temperature preference parameters set by the user, balance comfort and energy consumption through a multi-objective optimization algorithm; and generate a daily energy efficiency analysis report, marking the equipment and time periods with the greatest energy saving potential.

8. The method for dynamic load matching of low vacuum circulating water heating based on AI control according to claim 1 is characterized in that: It also includes regional coordinated control steps, including: Divide the building into multiple independent control zones, and set personalized temperature control strategies for each zone; Dynamically adjust the flow distribution of each branch through the hydraulic balance algorithm to eliminate uneven hot and cold pipe network; When it is detected that a certain area has low load for a long time, the corresponding branch will be automatically closed and anti-freeze protection will be activated.

9. The method for dynamic load matching of low vacuum circulating water heating based on AI control according to claim 1 is characterized in that: It also includes energy efficiency optimization steps: Store excess heat during off-peak hours in the grid, and dynamically adjust the amount of heat storage based on weather forecasts; Implement an intelligent rotation strategy for circulating pumps to balance the difference in cumulative operating time of each pump within 10%; A pressure fluctuation suppression algorithm is used to control the pressure difference fluctuation at key nodes of the pipeline network within the range of ±5%.

Images