Intermittent heating control method and system for residential building

Abstract

The invention discloses a residential building intermittent heating control method and system, and belongs to the technical field of building environment and energy application. The method comprises the following steps: acquiring a building envelope temperature sequence, indoor and outdoor environment data and weather forecast information through a data sensing module; the dynamic prediction module corrects weather forecast errors by using a long short-term memory (LSTM) network, constructs a multi-node thermal network model, and predicts an indoor temperature attenuation trend under a heating-free condition; the decision module evaluates the necessity of heating intervention based on the prediction result and dynamically optimizes the starting opportunity; and the optimization control module realizes adaptive optimization of model parameters through rolling prediction and deviation monitoring. The method effectively solves the problems that in the prior art, building thermal inertia quantization is insufficient, prediction deviation is large, and energy efficiency and comfort level are unbalanced, and has the advantages of being easy to implement, remarkable in energy-saving effect and wide in application range.

Description

Technical Field

[0001] This invention belongs to the field of building environment and energy application technology, specifically relating to a method and system for controlling intermittent heating in residential buildings based on meteorological forecast data and thermal inertia sensing. Background Technology

[0002] In hot-summer and cold-winter regions of my country, residential buildings primarily employ intermittent heating and individual heating systems in winter to balance indoor thermal comfort and energy conservation. Simultaneously, to meet building energy conservation and carbon reduction requirements, an increasing number of new buildings utilize improved insulation structures, leading to a significant increase in building thermal inertia. Increased thermal inertia results in greater temperature transfer attenuation and delay, while existing heating control strategies generally lack quantitative consideration of the impact of building thermal inertia.

[0003] Intermittent heating theoretically saves energy, but in practice, it presents several problems, primarily: 1) Intermittent heating relies heavily on fixed schedules or simple temperature threshold triggering mechanisms, failing to fully utilize the thermal inertia buffering effect of the building envelope. This leads to energy waste from premature heating or substandard comfort due to delayed start-up; 2) Insufficient preheating time prevents indoor temperatures from reaching the design temperature during building use, impacting user experience. To achieve rapid preheating, intermittent heating systems must increase their capacity to compensate for the slow preheating; 3) Existing forecasting technologies often use historical meteorological data or short-term forecasts as boundary conditions. However, the raw forecast data released by meteorological departments cannot meet the local building comfort forecasting requirements, and the forecasting models do not incorporate building thermal characteristics, resulting in insufficient adaptability to extreme weather conditions and an inability to accurately predict the indoor temperature decay trend over the next few hours to days; 4) Traditional thermal network model parameters are typically based on parameters given in the design drawings, failing to reflect dynamic factors such as material aging and changes in usage patterns during actual building operation. Static models struggle to track the time-varying characteristics of the thermal performance of the building envelope, leading to accumulated prediction biases and impacting control accuracy.

[0004] Currently, extensive research has been conducted on intermittent heating in China. Patent (CN 209688973 U) describes an intermittent heating system with an energy storage tank, addressing the drawbacks of intermittent heating by adding an energy storage device to the heating system. However, this intermittent heating mode with an energy storage tank cannot be applied to single-family residential heating due to space and cost factors. Patent (CN112097315 B) proposes a time-sharing and zone-based complementary intermittent heating method suitable for school district heating systems. It proposes using time-sharing and zone-based intermittent heating for different user levels, utilizing the complementarity of different user loads in time and space. This method is suitable for centralized heating systems in public buildings such as schools, commercial office areas, and enterprises during long holidays. Patent (CN 206001596U) describes a time-sharing and intermittent heating control device for public buildings. It transmits data from outdoor and indoor temperature sensors to the control center via a data transmission module, and the control center calculates PID control to adjust the valve opening.

[0005] Existing research on intermittent heating control in buildings mainly focuses on the heating sector of public buildings. However, the living characteristics, resident structure, and behavior of residents in residential buildings differ greatly from those in public buildings. The core objective of intermittent heating control in residential buildings is to ensure "targeted needs of residents" and "convenience of use for residents." It is difficult to simply apply the heating control model of public buildings to intermittent heating in residential buildings.

[0006] Based on the need to balance building thermal comfort and energy conservation and carbon reduction under the dual carbon objectives, there is an urgent need for a control method for intermittent heating systems that can integrate high-precision weather forecast correction, dynamic perception of building thermal inertia, and adaptive parameter optimization, in order to solve the bottleneck problem of existing technologies in balancing heating energy efficiency and comfort in residential buildings. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a method and system for controlling intermittent heating in residential buildings based on meteorological forecast data and thermal inertia sensing. By fusing data sensing (dynamic sensing of building thermal inertia) with weather forecast data, a Long Short-Term Memory (LSTM) network forecast error correction model is constructed, thereby achieving adaptive parameter optimization control of the intermittent heating system.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] The intermittent heating control system for residential buildings based on meteorological forecast data and thermal inertia sensing specifically includes: a data sensing module, a dynamic prediction module, a decision-making module, and an optimization control module.

[0010] A method for controlling intermittent heating in residential buildings based on meteorological forecast data and thermal inertia sensing includes the following steps:

[0011] S1. Data Sensing: Temperature measuring points are set up at key locations on the exterior walls and main floor slabs of the building with typical orientations to collect temperature sequences of the inner and outer surfaces of the walls and floors and the intermediate layers. Combined with indoor air temperature recorded by the building automation system, historical data of indoor and outdoor temperature and humidity, and meteorological forecast information for the next 120 hours issued by the meteorological department, a dynamic thermal characteristic database of the building is constructed.

[0012] S2. Dynamic Prediction: By constructing a Long Short-Term Memory (LSTM) network forecast error correction model, the hourly weather forecasts issued by the meteorological department for the next 120 hours are corrected in real time. Based on building structural parameters, a thermal network model including walls, floors, and indoor air is established. The equivalent heat capacity and heat transfer coefficient of each component are identified online, and the model parameters are updated to predict the indoor air temperature under unheated conditions.

[0013] S3. Make heating decisions: Using the detailed weather forecast for the next 120 hours as the external boundary condition, and combining the thermal network model, the indoor air temperature is simulated and predicted hourly to determine whether the indoor temperature has fallen below the minimum comfort temperature limit specified by the user, so as to achieve a scientific assessment of the necessity of heating intervention and dynamic optimization of the start-up timing of the intermittent heating system.

[0014] S4. Optimization Control: The optimization control module, based on rolling prediction and online deviation monitoring, adaptively optimizes the model and updates parameters for the next round of prediction tasks, enabling continuous evolution of model performance.

[0015] Preferably, in step S2, the dynamic prediction module processes meteorological data using a pre-trained Long Short-Term Memory (LSTM) network model, establishes a multi-node lumped-parameter thermal network model including walls, floors, and indoor air based on building structural parameters, and identifies the equivalent heat capacity and heat transfer coefficient of each component online using the recursive least squares method. The steps include:

[0016] S21. Data preparation: Collect historical weather forecast data (including temperature and relative humidity) and actual measurement data for the corresponding time periods to form a training dataset;

[0017] S22. Construct a lumped parameter thermal network model (8 resistors and 7 capacitors), including: three thermal nodes for the outer layer, insulation layer, and inner layer of the exterior wall; three thermal nodes for the upper layer, concrete layer, and lower layer of the floor slab; and indoor and outdoor air nodes.

[0018] S23. The nodes are connected by thermal conduction, convection, and long-wave radiation thermal resistance. The state equations are expressed as follows:

[0019]

[0020] in

[0021] Let K be the heat capacity of the k-th hot node; For inter-node thermal resistance; For indoor heat sources; T is the temperature.

[0022] S24. Solve the above differential equations numerically using the fourth-order Runge-Kutta method, with a time step of 60 seconds, and predict the indoor air temperature 24 hours in advance. The decay curve;

[0023] S25. The Recursive Least Squares (RLS) method is called every 30 minutes to update the model parameters.

[0024] Preferably, in step S21, the dynamic prediction module uses a pre-trained Long Short-Term Memory (LSTM) network model to process meteorological data. The calculation and model training steps include:

[0025] S211. Collect historical weather forecast data sequences This includes temperature, relative humidity, and the actual measurement data sequence for the corresponding time period. ,in h, with a time resolution of 1h;

[0026] Standardize the data: , ,in and These are the mean and standard deviation, respectively.

[0027] Construct input-output pairs: The input is a sequence of differences between the forecast and actual values ​​over the past 24 hours. ,in The output is the predicted value of the forecast error for the next 24 hours. ;

[0028] S212. A multi-layer LSTM network is used to learn the dynamic characteristics of the prediction error. The network structure includes:

[0029] Input layer: 24-dimensional (corresponding to a 24-hour historical difference sequence);

[0030] Hidden layers: 2 layers of LSTM units, each containing 64 neurons, using the activation functions Sigmoid and tanh;

[0031] Output layer: Fully connected layer with a dimension of 24 (for 24-hour error prediction), using a linear activation function to adapt to the regression task;

[0032] The model was trained using time-series cross-validation, with mean squared error (MSE) as the loss function, Adam as the optimizer, a learning rate of 0.001, ≥1000 training cycles, and a batch size of 32.

[0033] S213. For real-time acquired 120-hour weather forecast sequences The trained LSTM model is used to predict the prediction error. Generate corrected meteorological sequences The corrected meteorological data is integrated with the building thermal characteristics database to form a high-precision, multi-dimensional dynamic thermal characteristics database, which is used for subsequent thermal network modeling and prediction.

[0034] Preferably, in step S25, the model parameters are updated using the recursive least squares (RLS) method, specifically including the following steps:

[0035] S251 Constructing the regression vector It is composed of the temperature difference at each node;

[0036] S252 calculates the gain vector. ;

[0037] in:

[0038] Indicates the first The Kalman gain vector of the step is used to determine the correction weights for parameter estimates based on new observation data.

[0039] Indicates the first The covariance matrix of the step reflects the degree of uncertainty in the current parameter estimation.

[0040] Indicates the first The regression vector of the step.

[0041] S253 Updated Parameter Estimation ;

[0042] Indicates the first The step estimates the unknown parameter vector, including the key thermodynamic parameters to be identified.

[0043] Indicates the system at the 1st The actual output measurement value at any given time, in °C.

[0044] S254 Update the covariance matrix ;

[0045] S255 sets the forgetting factor This gives higher weight to recent data.

[0046] Preferably, in step S3, the decision-making module scientifically assesses the necessity of heating intervention and dynamically optimizes the optimal start-up time. Specific steps include:

[0047] S31. Obtain the hourly outdoor temperature sequence for the next 120 hours released by the meteorological department. The time resolution is 1 hour, covering the period from the current time. The continuous 120-hour interval starting from;

[0048] S32. Calculate the average outdoor temperature over the entire 120-hour forecast period. If the average outdoor temperature... If the temperature is low for an extended period, it is determined that the building envelope's own heat storage is insufficient to maintain basic thermal comfort requirements. The heating system should be activated immediately, with the heat source output power preset to 80% to 100% of its rated power.

[0049] S33. If This indicates that the current temperature drop is only a temporary fluctuation, and the system enters energy-saving assessment mode; at this time, the dynamic prediction module outputs the results for the next 24 hours (i.e., Indoor air temperature under conditions without active heating The change curve; extract the minimum value from the predicted curve using the following formula:

[0050]

[0051] judge Is it below the user-specified minimum comfort temperature?

[0052] like This indicates that even without starting the heating system, the indoor temperature will drop below the safe threshold, posing a risk of heat loss. The heating system should be started immediately.

[0053] like This indicates that the building's thermal inertia is sufficient to withstand the impact of this short-term cooling, so the heating equipment should be turned off or kept off, and the zero-energy insulation mode should be activated.

[0054] S34. All control commands are sent to the Building Automation System (BAS) to drive the heating equipment to perform corresponding operations; the system updates meteorological data and measured temperature every 60 minutes, dynamically refreshing the data. and .

[0055] Preferably, the S4 optimization control module performs adaptive optimization and parameter updates for the model, specifically including the following steps:

[0056] S41. The system operates at fixed time intervals. Continuously run modules S1 to S3 to acquire the latest outdoor meteorological data, indoor and building envelope measurement point temperatures, and update the initial state of the multi-node thermal network model;

[0057] S42. Simultaneously extract the actual outdoor temperature measurement sequence of the past 6 hours each time the S3 decision is executed. and indoor temperature measurement sequence And compared with the corrected forecast values ​​generated by the S211 LSTM model for the corresponding time period. Compare;

[0058] S43. Calculate the maximum absolute forecast deviation within this time period using the following formula:

[0059]

[0060] Set forecast deviation threshold (For example ).like If the forecast deviation is too large, the system will jump to S47; otherwise, continue to execute S44.

[0061] S44. Extract the measured indoor air temperature values ​​from the past N hours that were not affected by active heating. The corresponding time period is compared with the predicted value generated by the S2 thermal network model. By comparison, the maximum absolute prediction bias is calculated:

[0062]

[0063] S45. If (For example If the current thermal network model is significantly inaccurate, the online calibration procedure for the thermal network equivalent parameters is triggered, and S46 is executed; otherwise, the existing model parameters remain unchanged.

[0064] S46. Online calibration uses an incremental proportional correction method to adjust the equivalent heat capacity of the walls and floors. :

[0065]

[0066] in:

[0067] The equivalent heat capacity after updating the k-th hot node;

[0068] The equivalent heat capacity before the k-th hot node is updated;

[0069] The learning rate is set to a dynamic value, initially set to 0.1.

[0070] The reference temperature is used for normalization and is usually taken as... .

[0071] S47. If the forecast deviation is determined to be too large in S43, then suspend this thermal network parameter calibration. The system will then use the latest actual meteorological data. The corresponding original forecast data is added to the training dataset of the LSTM model, and the incremental training process of the LSTM model is started to ensure the forecast accuracy in the next running cycle. The calibrated model parameters are immediately written to the local database for the next round of prediction tasks, realizing the continuous evolution of model performance.

[0072] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0073] 1. Compared with heating modes that determine the start-up of the heating system based on the date or the indoor ambient temperature, this invention provides an adaptive parameter optimization control method for intermittent heating systems based on the thermal inertia sensing of individual residential buildings, meteorological forecast data, and incremental training of LSTM models. This method can better balance the thermal comfort and energy-saving effect of intermittent heating in residential buildings during winter.

[0074] 2. This method is easy to implement and cost-controllable. It only requires pre-installing temperature measuring points at key locations on the exterior walls and main floor slabs of the building in typical orientations during the construction phase. These points are used to collect temperature sequences of the inner and outer surfaces of the walls and floors, as well as the intermediate layers. The test data is used to analyze thermal inertia, and no large-scale modifications to the heating system are required.

[0075] 3. This method is particularly suitable for residential buildings in hot-summer and cold-winter regions, and is also applicable to intermittent heating control of residential buildings at the beginning and end of the heating season in cold regions, resulting in significant energy savings. It has a wide range of applications and is highly operable. Attached Figure Description

[0076] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0077] Figure 1 is a system module diagram of the present invention.

[0078] Figure 2 This is a flowchart of the construction method of the heating control system of the present invention.

[0079] Figure 3 This is a schematic diagram of the lumped-parameter thermal network model for residential buildings according to the present invention.

[0080] Figure 4 The flowchart for the optimized control module of this invention is shown below. Detailed Implementation

[0081] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0082] This invention provides a technical solution: an intermittent heating control system for residential buildings based on meteorological forecast data and thermal inertia sensing, such as... Figure 1 As shown, it includes a data perception module, a dynamic prediction module, a decision-making module, and an optimization control module.

[0083] Data sensing module: used to collect temperature data of the inner and outer surfaces and intermediate layers of residential building walls and floors, indoor air temperature, historical indoor and outdoor temperature and humidity data, and meteorological forecast information for the next 120 hours issued by the meteorological department.

[0084] Dynamic prediction module: Predicts indoor temperature of residential buildings under unheated conditions based on relevant information data.

[0085] Decision module: Decides whether to start the heating network system based on the forecast results.

[0086] Optimized control module: Based on whether the short-term prediction deviation reaches the threshold, corresponding intermittent heating control optimization is performed.

[0087] A method for controlling intermittent heating in residential buildings based on meteorological forecast data and thermal inertia sensing, such as Figure 2 As shown, the feature is that it includes the following steps:

[0088] S1. Install armored platinum resistance temperature measuring points (Pt100 or Pt1000) at key locations on the exterior walls and main floor slabs of the building in typical orientations, collect temperature sequences of the inner and outer surfaces of the walls and floors and the intermediate layers, and combine them with the indoor air temperature recorded by the building automation system, historical data of indoor and outdoor temperature and humidity, and the weather forecast information for the next 120 hours issued by the meteorological department to construct a dynamic thermal characteristic database of the building.

[0089] S11. The external wall measuring points are arranged in three layers: the outer surface, the midpoint of the insulation layer, and the inner surface; each measuring point uses an armored platinum resistance sensor (Pt100 or Pt1000), with a probe diameter ≤6mm, a length ≥50mm, and an IP68 protection rating and a stainless steel armored sleeve.

[0090] S12. The floor slab measuring points are also arranged in three layers at the top / bottom floor and the center of the concrete layer, and the sensor installation method is the same as that of the exterior wall;

[0091] S13. All sensors support three-wire connection, with a sampling frequency of 1Hz. Data is uploaded to the local edge controller via shielded cable or wireless ZigBee module.

[0092] S14. Synchronously access data such as indoor dry bulb temperature (suspended 1.2m above the ground), fresh air volume, and equipment operating status from the building automation system (BAS), and integrate 120-hour cold wave forecast information released by the meteorological department, including temperature, wind speed, sunshine, and relative humidity, to form a multi-dimensional dynamic thermal characteristic database.

[0093] S2. Construct a Long Short-Term Memory (LSTM) network forecast error correction model to correct hourly weather forecasts issued by meteorological departments for the next 120 hours in real time. Based on building structural parameters, establish a multi-node lumped parameter thermal network model including walls, floors, and indoor air. Use the recursive least squares method to identify the equivalent heat capacity and heat transfer coefficient of each component online and predict the indoor air temperature under no-heating conditions.

[0094] S21. Data preparation: Collect historical weather forecast data (including temperature and relative humidity) and actual measurement data for the corresponding time periods to form a training dataset;

[0095] S211. Data Preparation and Preprocessing:

[0096] Collect historical weather forecast data sequences This includes temperature, relative humidity, and the actual measurement data sequence for the corresponding time period. ,in h, with a time resolution of 1h;

[0097] Standardize the data: , ,in and These are the mean and standard deviation, respectively.

[0098] Construct input-output pairs: The input is a sequence of differences between the forecast and actual values ​​over the past 24 hours. ,in The output is the predicted value of the forecast error for the next 24 hours. ;

[0099] S212. LSTM module computation and model training:

[0100] The dynamic characteristics of the prediction error are learned using a multi-layer LSTM network. The network structure includes:

[0101] Input layer: 24-dimensional (corresponding to a 24-hour historical difference sequence);

[0102] Hidden layers: 2 layers of LSTM units, each containing 64 neurons, using the activation functions Sigmoid and tanh;

[0103] Output layer: Fully connected layer with a dimension of 24 (for 24-hour error prediction), using a linear activation function to adapt to the regression task;

[0104] LSTM cells at time steps The calculation process is as follows:

[0105] Forget Gate: Determines which information is discarded from the cell state; the calculation formula is as follows.

[0106]

[0107] in The sigmoid activation function is used. Here is the forget gate weight matrix. This is the hidden state from the previous moment. Enter the current time. To offset the forget gate;

[0108] Input gate: Determines which new information is stored in the cell state, and includes two parts:

[0109]

[0110]

[0111] in , The input gate weight matrix, , For the corresponding bias;

[0112] Cell status update:

[0113]

[0114] in Indicates element-wise multiplication;

[0115] Output gate:

[0116]

[0117]

[0118] in This is the output gate weight matrix. For output gate bias, Hide the current state;

[0119] The model was trained using time-series cross-validation, and the loss function was mean squared error (MSE).

[0120]

[0121] The optimizer uses Adam, the learning rate is set to 0.001, the training epochs are ≥1000, and the batch size is 32;

[0122] S213. Real-time correction and data fusion:

[0123] For real-time acquired 120-hour weather forecast sequences The trained LSTM model is used to predict the prediction error. Generate corrected meteorological sequences The corrected meteorological data is integrated with the building thermal characteristics database to form a high-precision, multi-dimensional dynamic thermal characteristics database, which is used for subsequent thermal network modeling and prediction.

[0124] S22. Construct a lumped parameter thermal network model as follows: Figure 3 As shown, R is thermal resistance, in K / W; C is heat capacity, in J / ℃; and T is temperature, in ℃. These are the corrected outdoor meteorological parameters. T2, T3, T4, T6, T7, and T8 represent the temperatures at the thermal nodes of the outer wall layer, concrete layer, inner layer, upper floor slab, concrete layer, and lower floor, respectively. C w1 C w2 C w3 C m C m1 C m2 C m3 These are the heat capacities of the outer layer of the exterior wall, the concrete layer, the inner layer, the indoor air, the upper layer of the floor slab, the concrete layer, and the lower layer of the thermal nodes. For example... Figure 3 The heat transfer network shown includes three thermal nodes: the outer layer of the exterior wall, the insulation layer, and the inner layer; three thermal nodes: the upper layer of the floor slab, the concrete layer, and the lower layer; and indoor and outdoor air nodes.

[0125] S23. The nodes are connected by thermal conduction, convection, and long-wave radiation thermal resistance. The state equations are expressed as follows:

[0126]

[0127] in

[0128] Let K be the heat capacity of the k-th hot node;

[0129] For inter-node thermal resistance;

[0130] For indoor heat sources;

[0131] S24. Solve the above differential equations numerically using the fourth-order Runge-Kutta method (RK4), with a time step of 60 seconds, and predict the indoor air temperature over the next 48 hours. The decay curve;

[0132] S25. Update the model parameters using Recursive Least Squares (RLS) every 30 minutes. The specific steps are as follows:

[0133] - Constructing the regression vector It is composed of the temperature difference at each node;

[0134] - Calculate the gain vector ;

[0135] - Update parameter estimates ;

[0136] - Update covariance matrix ;

[0137] - Set the forgetting factor This gives higher weight to recent data.

[0138] in:

[0139] Indicates the first The Kalman gain vector of the step is used to determine the correction weights for parameter estimates based on new observation data.

[0140] Indicates the first The covariance matrix of the step reflects the degree of uncertainty in the current parameter estimation. The initial values ​​are set to... ,in A large positive number (such as) ), indicating that the initial parameters are unknown.

[0141] Indicates the first The regression vector of the step.

[0142] This is the forgetting factor, with a value ranging from 0.95 to 0.99.

[0143] Indicates the first The step estimates the unknown parameter vector, including the key thermodynamic parameters to be identified.

[0144] Indicates the system at the 1st The actual output measurement value at any given time, in °C.

[0145] Updated parameters output by S25 It is stored in the local model database and used to solve the differential equations of S24.

[0146] S3 uses a detailed 120-hour weather forecast as the external boundary condition and combines a multi-node lumped parameter thermal network model to perform full-time simulation and prediction of the indoor air temperature evolution process. It determines whether the indoor temperature falls below the user's minimum comfort temperature limit (18℃), thereby achieving a scientific assessment of the necessity of heating intervention and dynamic optimization of the optimal start-up time.

[0147] S31. Obtain the hourly outdoor temperature sequence for the next 120 hours released by the meteorological department. The time resolution is 1 hour, covering the period from the current time. The continuous 120-hour interval starting from;

[0148] S32. Calculate the average outdoor temperature over the entire 120-hour forecast period:

[0149]

[0150] like If the temperature is low for an extended period, it is determined that the building envelope's own heat storage is insufficient to maintain basic thermal comfort requirements. The heating system should be activated immediately, with the heat source output power preset to 80% to 100% of its rated power.

[0151] S33. If This indicates that the current temperature drop is only a temporary fluctuation, and the system enters the energy-saving assessment mode. At this time, the multi-node lumped parameter thermal network model established by S2 is invoked, using the current temperature of each measuring point on the wall and floor as the initial condition, and the corrected future 24-hour temperature as the starting point. As a boundary input, the energy conservation equations are numerically solved to predict the next 24 hours (i.e., ...). Indoor air temperature under conditions without active heating Change curve;

[0152] S34. Extract the minimum value from the predicted curve:

[0153]

[0154] S35. Judgment Is it below the user-specified minimum comfort temperature (assuming)? ):

[0155] like This indicates that even without starting the heating system, the indoor temperature will drop below the safe threshold, posing a risk of heat loss. The heating system should be started immediately.

[0156] like This indicates that the building's thermal inertia is sufficient to withstand the impact of this short-term cooling, so the heating equipment should be turned off or kept off, and the zero-energy insulation mode should be activated.

[0157] S36. All control commands are sent to the Building Automation System (BAS) to drive the heating equipment to perform corresponding operations; the system updates meteorological data and measured temperature every 60 minutes, dynamically refreshing the data. and .

[0158] S4. Model adaptive optimization decision-making process based on rolling forecasting and online deviation monitoring, such as... Figure 4 As shown.

[0159] S41. The system operates at fixed time intervals. Continuously run modules S1 to S3 to acquire the latest outdoor meteorological data, indoor and building envelope measurement point temperatures, and update the initial state of the multi-node thermal network model;

[0160] S42. Simultaneously extract the actual outdoor temperature measurement sequence of the past 6 hours each time the S3 decision is executed. and indoor temperature measurement sequence And compared with the corrected forecast values ​​generated by the LSTM model of S211 for the corresponding time period. Comparison; revised forecast values ​​are derived from... Figure 3 The computational grid is obtained.

[0161] S43. Calculate the maximum absolute forecast deviation within this time period:

[0162]

[0163] Set forecast deviation threshold (For example ).like If the forecast deviation is too large, the system will jump to S47; otherwise, continue to execute S44.

[0164] S44. Extract the measured indoor air temperature values ​​that were not affected by active heating over the past N hours. The corresponding time period is compared with the predicted value generated by the S2 thermal network model. By comparison, the maximum absolute prediction bias is calculated:

[0165]

[0166] S45. If (For example If the current thermal network model is significantly inaccurate, the online calibration procedure for the thermal network equivalent parameters is triggered, and S46 is executed; otherwise, the existing model parameters remain unchanged.

[0167] S46. Online calibration uses an incremental proportional correction method to adjust the equivalent heat capacity of the walls and floors. :

[0168]

[0169] in:

[0170] The equivalent heat capacity after updating the k-th hot node;

[0171] The equivalent heat capacity before the k-th hot node is updated;

[0172] The learning rate is set to a dynamic value, initially set to 0.1.

[0173] The reference temperature is used for normalization and is usually taken as... .

[0174] When the system detects M consecutive calibrations (e.g., M=3), If no decreasing trend is observed, the learning rate will decay to [a certain value]. To prevent overcorrection and oscillation;

[0175] S47. If the forecast deviation is determined to be too large in S43, then suspend this thermal network parameter calibration. The system will then use the latest actual meteorological data. The corresponding original forecast data is added to the training dataset of the LSTM model, and the incremental training process of the LSTM model is started to ensure the forecast accuracy in the next running cycle. The calibrated model parameters are immediately written to the local database for the next round of prediction tasks, realizing the continuous evolution of model performance.

Claims

1. A method for controlling intermittent heating of a residential building, characterized by, The method comprises the following steps: Step S1. Data sensing: temperature measuring points are arranged at key positions of the building's typical orientation outer wall and floor slab, temperature sequences of the inner and outer surfaces of the wall and floor slab and the intermediate layer are collected, indoor air temperature recorded by the building automation system, indoor and outdoor temperature and humidity historical data, and future 120h weather forecast information released by the meteorological department are combined to construct a building dynamic thermal characteristic database; Step S2. Dynamic prediction: the future 120h hourly weather forecast released by the meteorological department is corrected in real time by constructing a long short-term memory network LSTM prediction error correction model; a thermal network model including walls, floors and indoor air is established based on building structure parameters, the equivalent heat capacity and heat transfer coefficient of each component are identified online and the model parameters are updated, and the indoor air temperature under the condition of no heating is predicted; Step S3. Heating decision: the future 120h refined weather forecast is taken as the external boundary condition, the indoor air temperature is simulated and predicted hour by hour combined with the thermal network model, it is judged whether the indoor temperature drops below the minimum comfort temperature limit value specified by the user, and the dynamic optimization of heating intervention is realized; Step S4. Optimization control: the optimization control module based on rolling prediction and online deviation monitoring is used for model adaptive optimization and parameter updating for the next round of prediction task, and the continuous evolution of the model performance is realized.

2. The intermittent heating control method for a residential building according to claim 1, characterized by: The S2 comprises the following steps: S21. Data preparation: collect historical weather forecast data and actual measurement data in the corresponding period to form a training data set; S22. Construct a centralized parameter thermal network model, including: three thermal nodes of the outer layer of the outer wall, the insulation layer and the inner layer; Three thermal nodes of the upper layer of the floor, the concrete layer and the lower layer; indoor and outdoor air nodes; S23. The nodes are connected through thermal conductivity, convection and long-wave radiation resistance, and the state equation set is represented as: ，k=1，2，3； Wherein Ck is the heat capacity for the kth thermal node; Rth is the thermal resistance between nodes; is the indoor heat source term; S24. Numerical solution of the above system of differential equations using the fourth order Runge-Kutta method RK4 with a time step of 60 seconds, predicting the decay curve of the indoor air temperature for 24 hours forward in time; S25. Recursive least squares RLS is called once every 30 minutes to update the model parameters.

3. The intermittent heating control method for residential buildings according to claim 1, characterized in that: The S3 comprises the following steps: S31. Obtain a future 120h hourly outdoor air temperature sequence with a time resolution of 1h covering a continuous 120h interval from the current time onwards; S32. Calculate the average outdoor air temperature in the whole 120h forecast period, if the average outdoor air temperature is less than 0℃, then determine it as a long-term persistent low temperature process, the building envelope itself cannot maintain the basic thermal comfort demand by heat storage, immediately start the heating system, and preset the heat source output power as 80%~100% of the rated power; S33. If , it indicates that this cooling is only a phase fluctuation, and enters the energy-saving evaluation mode; the future 24h output of the dynamic prediction module , the predicted curve of the change of indoor air temperature under the condition of no active heating ; extract the minimum value in the predicted curve, as follows:

4. determining whether it is below a minimum comfort temperature specified by the user ; If then it is indicated that even if the heating is not activated, the indoor temperature will fall below the safety threshold and there is a risk of hypothermia, the heating system is activated immediately; If , it indicates that the building's thermal inertia is sufficient to resist the impact of this round of short-term cooling, and the heating equipment is closed or kept closed, and the zero-energy insulation mode is enabled. S34. All control instructions are issued to the building automation system BAS to drive the heating equipment to perform corresponding operations; the system updates the weather data and measured temperature every 60 minutes to dynamically refresh with .

5. The intermittent heating control method for a residential building according to claim 1, characterized by: The S4 comprises the following steps: S41. The system updates the initial state of the multi-node thermal network model at fixed time intervals The S1-S3 modules are continuously running to obtain the latest outdoor meteorological data, indoor and building envelope temperature measurement points, and update the initial state of the multi-node thermal network model. S42. At each execution of the S3 decision, synchronously extract the actual outdoor temperature measurement sequence of the past 6h and the indoor temperature measurement sequence and compare them to the corrected forecast values generated by the LSTM forecast error correction model for the corresponding time period ; S43. Calculate the maximum absolute prediction deviation in this period, and the specific formula is as follows: ； Setting a prediction deviation threshold If , then determine the current prediction deviation, jump to S47; otherwise, continue to execute S44; S44. Extracting measured values of indoor air temperature not affected by active heating in the past Nh Comparing with the predicted values generated by the S2 thermal network model for the corresponding period Comparing, calculating the maximum absolute prediction error: ； S45. If a significant misalignment is detected, the online calibration procedure of the thermal network equivalent parameters is triggered and S46 is executed; otherwise, the existing model parameters are maintained. S46. Online calibration employs incremental proportional correction method to adjust the equivalent heat capacity of walls and floors : ； Wherein: updated equivalent thermal capacitance for the kth thermal node; Equivalent thermal capacitance before update for the kth thermal node; The initial value is set to 0.1 for the dynamic learning rate. For reference temperature, for normalization, usually taken ; S47. If it is determined in S43 that the forecast deviation is too large, suspend the current heat network parameter calibration; the system adds the latest actual weather data corresponding to the original forecast data to the training data set of the LSTM model, and starts the incremental training process of the LSTM model to ensure its forecast correction accuracy in the next operation cycle; The calibrated model parameters are immediately written into the local database for the next round of prediction task, and the continuous evolution of the model performance is realized.

6. The intermittent heating control method for a residential building according to claim 2, characterized by: The S21 comprises the following steps: S211. Collecting a sequence of historical weather forecast data , including temperature, relative humidity, and a sequence of actual measurement data for the corresponding time period wherein h, a time resolution of 1 h; Standardizing the data: , where and are the mean and standard deviation, respectively. Construct input-output pairs: input is the sequence of differences between forecast and actual values for the past 24h where output is the forecast of the forecast error for the next 24h ; S212. Use a multi-layer LSTM network to learn the dynamic characteristics of the prediction error, and the network structure comprises: Input layer: dimension 24; Hidden layer: 2 layers of LSTM units, each layer contains 64 neurons, and uses the activation functions Sigmoid and tanh; Output layer: fully connected layer, dimension 24, using linear activation function to adapt to regression task; The model training adopts time series cross-validation, the loss function is mean square error MSE, the optimizer uses Adam, the learning rate is set to 0.001, the training period is ≥1000 times, and the batch size is 32; S213. For the real-time acquired future 120h weather forecast sequence , using the trained LSTM model to predict the forecast error ; generate the corrected weather sequence ; fuse the corrected weather data with the building thermal characteristic database to form a high-precision multi-dimensional dynamic thermal characteristic database for subsequent thermal network modeling and prediction.

7. The intermittent heating control method for a residential building according to claim 2, characterized by: The S25 comprises the following steps: S251 constructs a regression vector comprising temperature differences of the nodes S252 compute gain vector ; Wherein: represents the Kalman gain vector of the step for determining the correction weight of the new observation data to the parameter estimation. represents the covariance matrix of the step, reflecting the degree of uncertainty of the current parameter estimate; represents the step of regression vector; S253 updating parameter estimates ; represents the first step estimates the value of the unknown parameter vector, containing the key thermodynamic parameters to be identified; the actual output measurement of the system at the first time instant, in °C; and the actual output measurement of the system at the second time instant, in °C. S254 update covariance matrix ; S255 sets the forgetting factor , giving higher weight to more recent data.

Images