Heat pump type secondary side water temperature closed loop regulation method and system

Abstract

The present application relates to the technical field of temperature closed-loop regulation, and discloses a heat pump type secondary side water temperature closed-loop regulation method and system, which comprises the following steps: obtaining historical load characteristic data of a target area, training a recurrent neural network model, obtaining a predicted load rate curve and an uncertainty range based on the trained recurrent neural network model, generating an automatic regulation strategy of a primary side valve opening degree, a secondary side circulating water pump frequency and a heat pump water supply temperature set value, and issuing the automatic regulation strategy to a controller. The present application can identify the future load change trend in advance before heating control by constructing a recurrent neural network model facing the future period, and generate an automatic regulation strategy in combination with the predicted load rate curve and the uncertainty range, thereby changing the control mode of the prior art which mainly relies on passive feedback regulation of the current working condition.

Description

Technical Field

[0001] This invention relates to the field of temperature closed-loop control technology, specifically to a method and system for closed-loop temperature control of heat pump secondary side water supply. Background Technology

[0002] Plate heat exchangers are key heat exchange and control equipment in centralized heating systems. Their operating status directly affects heating quality, system energy consumption, and user comfort. Existing heating control methods mainly rely on empirical setpoints, fixed control curves, or feedback regulation based on current operating conditions. They typically use single-parameter or few-parameter control based on primary-side valve opening, secondary-side circulating water pump frequency, or supply water temperature. While these control methods can meet the operating requirements under normal conditions, they struggle to develop control strategies adapted to future loads under continuously fluctuating outdoor weather conditions, especially in scenarios with large temperature differences between day and night, sudden cold waves, or short-term changes in heating demand in localized building units. This can easily lead to problems such as untimely heating response, insufficient heating in some areas, or excessive heating.

[0003] On the other hand, existing heating regulation typically focuses more on unified control at the overall unit level, with weak compatibility with personalized adjustment needs on the user side. While some systems allow users to make local adjustments via terminal valves, these adjustments are mostly independent of the unit's automatic control logic, lacking a linkage mechanism with load rate prediction results and primary and secondary control parameters. When users manually intervene in local adjustment devices such as terminal units, existing technologies often cannot accurately assess the impact of this adjustment on local flow distribution, additional heat demand, and the overall heat output of the unit, nor can they synchronously correct the primary valve opening based on this impact. This easily leads to a disconnect between local personalized control and global automatic control. This not only affects the temperature control accuracy of the target area but may also cause imbalances in system heat distribution, increased control fluctuations, and increased energy consumption. Therefore, achieving coordinated linkage between local personalized control and global heating regulation has become a pressing technical problem to be solved in this field. Summary of the Invention

[0004] To address the problems of lagging load change response, lack of forward-looking control methods, and insufficient coordination between user-side regulation and unit-side control in existing plate heat exchanger units, this invention provides a closed-loop regulation method and system for the secondary side water supply temperature of a heat pump.

[0005] The present invention adopts the following technical solution:

[0006] A closed-loop regulation method for secondary-side water supply temperature in a heat pump system is applied to a central hot water heating system comprising a heat pump unit, a plate heat exchanger unit, a primary-side circulation loop, a secondary-side circulation loop, and terminal devices, including:

[0007] Acquire historical load characteristic data for the target area, including historical meteorological data, user behavior data, building thermal inertia parameters, and historical heat exchanger unit load rates;

[0008] The acquired historical load characteristic data is analyzed, a recurrent neural network model is trained, and the load rate prediction sequence of the heat exchanger unit at time k is predicted.

[0009] Based on the trained recurrent neural network model, the predicted load rate curve and uncertainty range are obtained;

[0010] An automatic control strategy is generated based on the predicted load rate curve and the uncertainty range to control the primary valve opening, secondary circulating water pump frequency and heat pump supply temperature setpoint, and then sent to the controller.

[0011] Receive the terminal device adjustment command, integrate the terminal device adjustment command with the automatic control strategy, execute it according to the priority rules, and obtain the final execution opening of the terminal device at the current time step;

[0012] Based on the final execution opening, the primary valve opening fine-tuning amount is generated, and the updated primary valve opening, secondary circulating water pump frequency, heat pump water supply temperature setpoint, and final execution opening of the terminal device are jointly sent to the controller for execution.

[0013] As a further description of the above technical solution: the method for predicting the load rate of the heat exchanger unit at time k in the future includes:

[0014] Based on the historical load characteristic data, a training sample set is constructed using a sliding window approach;

[0015] The multi-source feature vectors of T consecutive historical moments are used as input samples, labeled as the heat exchanger unit load rate sequence at the next k moments; the sliding window step size is s hours, and multiple training samples are generated sequentially; the training samples are divided into training set, validation set and test set;

[0016] The training samples are input into a recurrent neural network model for training. The recurrent neural network model includes one or two recurrent layers, each containing a preferred range of 32 to 128 hidden units, and outputs a load rate prediction sequence for the future k time step through a fully connected layer. The network uses ReLU or tanh activation functions, Adam optimizer, and a learning rate range of 0.001 to 0.01. The training batch size is 32 to 128. During training, the mean squared error is minimized as the loss function.

[0017] As a further description of the above technical solution: the method for obtaining the predicted load factor curve includes:

[0018] Using a multi-source feature vector sequence of T consecutive historical moments as the input sequence of a recurrent neural network model, the input sequence is fed into the trained recurrent neural network model to obtain a load rate prediction sequence for the next k time steps.

[0019] The input sequence is introduced with random perturbation and Monte Carlo sampling to generate multiple sets of perturbation input sequences. Each set of perturbation input sequences is then input into the recurrent neural network model to obtain multiple sets of load rate prediction results corresponding to the next k time steps.

[0020] The average load rate of multiple load rate prediction results corresponding to each time step is calculated and used as the predicted load rate point value for that time step. The average load rate of each time step is connected to form a load rate prediction curve.

[0021] As a further description of the above technical solution: the method for obtaining the uncertainty range includes:

[0022] The standard deviation is calculated for the predicted sample sequence at each time step to quantify the fluctuation range of the predicted value at that time step.

[0023] Based on the standard deviation and a pre-set confidence coefficient, the upper and lower bounds of the confidence interval for each time step are calculated.

[0024] The load rate average at each time step is connected to form a load rate prediction curve, and the corresponding upper and lower confidence intervals are output as the uncertainty range.

[0025] As a further description of the above technical solution: the method for obtaining the primary valve opening degree and the secondary circulating water pump frequency is as follows:

[0026] Obtain the mean load rate and the upper bound of the confidence interval for the current time step, i.e., the upper limit of the predicted load rate;

[0027] Multiply the upper limit of the predicted load rate by the rated heat exchange capacity of the heat exchange unit to convert it into the heat required on the primary side;

[0028] Calculate the required primary flow rate at each time step based on the heat required on the primary side, and map the required primary flow rate to the primary valve opening.

[0029] The obtained primary valve opening is subject to valve opening limit and smoothing limit to generate the final primary valve opening;

[0030] The heat required on the primary side is taken as the heat required on the secondary side. The secondary side circulating water flow rate is calculated using the pipeline heat balance formula. The calculated secondary side circulating water flow rate is mapped to the pump drive frequency using the linear proportional relationship of pump characteristics.

[0031] As a further description of the above technical solution: the method for generating the final primary valve opening includes:

[0032] For each time step, the calculated valve opening is checked to see if it exceeds the range of the valve's minimum and maximum allowable opening. If the calculated value is lower than the valve's minimum allowable opening, the valve opening is set to the minimum allowable opening; if the calculated value is higher than the valve's maximum allowable opening, the valve opening is set to the maximum allowable opening.

[0033] The maximum variation range of the valve opening is preset. The difference between the valve opening at the current time step and the valve opening at the previous time step must not exceed the preset maximum variation range. If it exceeds the maximum variation range, a valve opening variation constraint interval is constructed at the current time step according to the preset maximum variation range. The primary valve opening is then corrected to the boundary value corresponding to the variation constraint interval to obtain the final primary valve opening at the current time step.

[0034] As a further description of the above technical solution: the method for obtaining the heat pump water supply temperature setpoint is as follows:

[0035] The acquired primary heat requirement, secondary circulating water flow rate, building thermal inertia parameters, and user behavior data are input into a pre-built water supply temperature prediction model, which outputs the heat pump water supply temperature setpoint for the current time step.

[0036] As a further description of the above technical solution: the method of integrating the end-device adjustment command with the automatic control strategy and executing it according to priority rules to obtain the final execution opening degree of the end-device at the current time step includes:

[0037] The system receives terminal device adjustment instructions sent by users through a remote APP platform. Each terminal device adjustment instruction is accompanied by a timestamp and a corresponding building unit identifier. The system then converts the received terminal device adjustment instructions into standardized control data.

[0038] Based on the current time step, extract the local control parameters corresponding to the building unit where the target terminal device is located from the automatic control strategy that has been issued to the controller;

[0039] Map the user-sent end-device adjustment commands to the corresponding opening correction values;

[0040] Establish a priority rule between the end device adjustment command and the automatic control strategy, and fuse the opening correction amount and the automatic control strategy according to the priority rule to obtain the final execution opening of the end device at the current time step.

[0041] As a further description of the above technical solution: the local control parameters include: the opening degree of the primary valve, the frequency of the secondary circulating water pump, the set value of the heat pump water supply temperature, and the basic opening degree of the target terminal device at the current time step.

[0042] As a further description of the above technical solution: the method for generating the primary valve opening fine-tuning amount based on the final execution opening degree includes:

[0043] Obtain the final terminal device opening offset and convert it to obtain the local branch flow offset;

[0044] The local branch flow offset, the secondary side supply and return water temperature difference at the current time step, and the building thermal inertia parameters of the building unit corresponding to the terminal device are input into the pre-built offset prediction model, and the additional heat demand offset is output.

[0045] If the offset of additional heat demand does not exceed the preset correction threshold, the current automatic control strategy will remain unchanged, and only the local offset control of the terminal device will be executed.

[0046] If the preset correction threshold is exceeded, the heat compensation ratio is obtained based on the ratio of the additional heat demand offset to the current rated heat exchange capacity of the heat exchange unit. Combined with the primary side valve opening characteristic curve, the heat compensation ratio is mapped to the primary side valve opening fine adjustment amount.

[0047] A heat pump-type secondary side water supply temperature closed-loop regulation system, used to implement the aforementioned heat pump-type secondary side water supply temperature closed-loop regulation method, includes:

[0048] Historical data acquisition module: used to acquire historical load characteristic data of the target area, including historical meteorological data, user behavior data, building thermal inertia parameters and historical heat exchanger unit load rate;

[0049] Model training module: used to analyze the acquired historical load characteristic data, train the recurrent neural network model, and predict the load rate prediction sequence of the heat exchanger unit at time k in the future;

[0050] Load rate prediction module: used to obtain the predicted load rate curve and uncertainty range based on the trained recurrent neural network model;

[0051] Strategy generation module: Used to generate automatic control strategies for primary valve opening, secondary circulating water pump frequency and heat pump supply temperature setpoints based on the predicted load rate curve and uncertainty range, and send them to the controller;

[0052] Data fusion module: Used to receive adjustment instructions from the end device, fuse the adjustment instructions with the automatic control strategy, execute according to priority rules, and obtain the final execution opening degree of the end device at the current time step;

[0053] Fine-tuning execution module: Used to generate a fine-tuning amount for the primary valve opening based on the final execution opening, and send the updated primary valve opening, secondary circulating water pump frequency, heat pump supply water temperature setpoint, and final execution opening of the terminal device to the controller for execution.

[0054] The beneficial effects of this invention are as follows:

[0055] This invention acquires historical meteorological data, user behavior data, building thermal inertia parameters, and historical heat exchanger unit load rates for the target area. It then constructs a recurrent neural network model oriented towards future time periods. This model can identify future load change trends before heating control begins and, combined with the predicted load rate curve and uncertainty range, generates an automatic control strategy for primary-side valve opening, secondary-side circulating water pump frequency, and heat pump supply water temperature setpoints. This changes the existing control method, which mainly relies on passive feedback adjustment based on current operating conditions. Through this predictive-driven control mechanism, the heat exchanger unit's ability to respond to load changes in advance is improved, making the heating output more aligned with future actual needs. This reduces problems of insufficient or excessive heating caused by control lag, and improves heating stability and control accuracy.

[0056] Furthermore, this invention does not treat user-side terminal device adjustment as an isolated operation separate from automatic control. Instead, it integrates terminal device adjustment commands with automatic control strategies according to priority rules, and performs coordinated fine-tuning of primary-side valve openings based on the final execution opening degree of the terminal device, local branch flow offsets, and additional heat demand offsets. This allows personalized adjustment needs in local areas to be incorporated into the overall unit control chain, ensuring consistency between local manual intervention and automatic control based on load rate prediction, and preventing disorderly impacts on the overall heat distribution of the system caused by local adjustments. This technical solution improves both the personalized heating control capability of the target area and enhances the coordination and operational stability of the entire heating system under complex conditions, thereby achieving a balance between improved heating quality and optimized system operating efficiency. Attached Figure Description

[0057] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0058] Figure 1 A flowchart of the heat pump-type secondary side water supply temperature closed-loop regulation method provided in Embodiment 1 of the present invention;

[0059] Figure 2 Here is a flowchart of the method for obtaining the predicted load factor curve provided in Embodiment 1 of the present invention;

[0060] Figure 3 This is a flowchart of the method for obtaining the uncertainty range provided in Embodiment 1 of the present invention;

[0061] Figure 4 The flowchart is for the heat pump type secondary side water supply temperature closed-loop regulation system provided in Embodiment 2 of the present invention. Detailed Implementation

[0062] To make the technical means, creative features, objectives, and effects of this invention readily understandable, the invention is further described below with reference to specific illustrations. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0063] Example 1

[0064] Please see Figures 1-3 This invention provides a technical solution: a closed-loop regulation method for the secondary side water supply temperature of a heat pump, applied to a central hot water heating system including a heat pump unit, a plate heat exchanger unit, a primary side circulation loop, a secondary side circulation loop, and terminal devices, comprising:

[0065] Acquire historical load characteristic data for the target area, including historical meteorological data, user behavior data, building thermal inertia parameters, and historical heat exchanger unit load rates;

[0066] The method for obtaining historical meteorological data includes: obtaining historical meteorological data of the target area for at least one year through meteorological departments, third-party meteorological service platforms and on-site micro-meteorological stations, including temperature, humidity, wind speed, wind direction, solar radiation and precipitation. When obtaining historical meteorological data, it is ensured that the data timestamp and regional coordinates correspond accurately.

[0067] The historical heat exchanger load rate is obtained by inverting historical operating parameters, acquiring the actual heat exchange of the heat exchanger at each historical moment, and then calculating the ratio with the rated heat exchange capacity.

[0068] It should be noted that the actual heat exchange is the heat actually transferred to the secondary side pipeline network by the heat exchanger unit during operation, calculated from the flow rate and temperature difference. The calculation formula is as follows: ; For actual heat exchange, For traffic, For the specific heat capacity of water, This refers to the temperature difference between the inlet and outlet water on the primary side.

[0069] The system acquires the user's indoor temperature through a smart temperature control terminal or indoor environment monitoring module, anonymizes the user's indoor temperature data, and maintains the matching of timestamps and spatial coordinates with building units to obtain the average temperature and temperature difference of the target area, forming user behavior data that can be used for load factor prediction.

[0070] The building thermal inertia parameters include the heat capacity, thermal conductivity, and effective area of ​​the building walls, roof, floor, and windows in the heating area.

[0071] The acquired historical load characteristic data is analyzed, a recurrent neural network model is trained, and the load rate prediction sequence of the heat exchanger unit at time k is predicted.

[0072] Methods for predicting the load rate of heat exchanger units at time k include:

[0073] Based on the historical load characteristic data, a training sample set is constructed using a sliding window approach;

[0074] Specifically, the multi-source feature vectors of T consecutive historical moments are used as input samples, and the labels are the heat exchanger unit load rate sequence at the next k moments; the sliding window step size is s hours, and multiple training samples are generated in sequence; the training samples are divided into training set, validation set and test set, with the preferred ratio being 7:2:1;

[0075] Optionally, the historical window length T = 24 hours; the prediction step size k = 6 hours; and the sliding window step size s = 1 hour.

[0076] First training sample input: ;in This represents the historical load characteristic data for hour t; t∈[1,24];

[0077] The corresponding tag is the load rate for the next 6 hours: ; This is the load rate for the 25th hour, which is the load rate for the next hour.

[0078] The sliding window generates the next sample, sliding every hour s=1. The second training sample is:

[0079] ; ;

[0080] This process continues until the end of the data sequence, forming the training samples.

[0081] The training samples are input into a recurrent neural network (RNN) model for training. The RNN model is a Long Short-Term Memory (LSTM) network or a gated recurrent unit (GRU), comprising one or two recurrent layers, each containing a preferred range of 32 to 128 hidden units. It outputs a predicted load rate sequence for the next k time steps through a fully connected layer. The network uses ReLU or tanh activation functions, employs Adam as the optimizer, and has a learning rate selectable from 0.001 to 0.01. The training batch size is 32 to 128. During training, minimizing the mean squared error is used as the loss function.

[0082] The loss function is: In the formula, The total number of training samples, Let i be the actual load rate of the i-th sample at time j. Let be the predicted load rate of the i-th sample at time j.

[0083] Based on the trained recurrent neural network model, the predicted load rate curve and uncertainty range are obtained;

[0084] Methods for obtaining the predicted load factor curve include:

[0085] Using a multi-source feature vector sequence of T consecutive historical moments as the input sequence of a recurrent neural network model, the input sequence is fed into the trained recurrent neural network model to obtain a load rate prediction sequence for the next k time steps.

[0086] The input sequence is introduced with random perturbation and Monte Carlo sampling to generate multiple sets of perturbation input sequences. Each set of perturbation input sequences is then input into the recurrent neural network model to obtain multiple sets of load rate prediction results corresponding to the next k time steps. The multiple sets of load rate prediction results corresponding to the same future time step constitute the prediction sample sequence for that time step.

[0087] It should be noted that the Monte Carlo sampling process refers to: adding random perturbations to the input of each time step, such as simulating temperature or load measurement errors, inputting the perturbations into the recurrent neural network model multiple times to obtain multiple sets of prediction outputs. Each set of outputs represents the load rate that may occur under such perturbations, and the combination of multiple sets of outputs forms the prediction sample sequence for that time step.

[0088] The average load rate of multiple load rate prediction results corresponding to each time step is calculated and used as the predicted load rate point value for that time step. The average load rate of each time step is connected to form a load rate prediction curve.

[0089] Methods for obtaining the range of uncertainty include:

[0090] The standard deviation is calculated for the predicted sample sequence at each time step to quantify the fluctuation range of the predicted value at that time step.

[0091] Based on the standard deviation and combined with a pre-set confidence coefficient, such as 95% or 90%, calculate the upper and lower bounds of the confidence interval for each time step;

[0092] The load rate average at each time step is connected to form a complete load rate prediction curve, and the corresponding upper and lower confidence intervals are output as the uncertainty range.

[0093] In some implementations, to illustrate the generation process of the predicted load factor curve and uncertainty range, for example, at a certain prediction time step, 10 sets of predicted load factor values ​​are obtained through Monte Carlo sampling, namely 0.68, 0.70, 0.71, 0.69, 0.72, 0.73, 0.70, 0.69, 0.71, and 0.72. The mean of these 10 sets of predicted load factor values ​​is calculated, yielding a predicted load factor point value of approximately 0.71 for the current prediction time step. Further calculation of the standard deviation of these 10 sets of predicted load factor values ​​yields the fluctuation range corresponding to the current prediction time step. Combined with the currently used preset confidence coefficient, the lower bound of the confidence interval for the current prediction time step is approximately 0.68, and the upper bound of the confidence interval is approximately 0.74. Therefore, 0.71 is used as the predicted load factor point value for the current prediction time step, 0.68 to 0.74 is used as the uncertainty range corresponding to the current prediction time step, and the upper bound of the confidence interval is used as the input for subsequent calculations of the required heat on the primary side.

[0094] Based on the predicted load rate curve and uncertainty range, an automatic control strategy is generated, which includes the primary side valve opening, the secondary side circulating water pump frequency, and the heat pump supply water temperature setpoint, and then sent to the controller.

[0095] The method for obtaining the primary valve opening and the secondary circulating water pump frequency is as follows:

[0096] Obtain the mean load factor predicted at the current time step and the upper bound of the confidence interval, i.e., the upper limit of the predicted load factor. In the formula, This is the upper bound of the confidence interval. To predict the average load factor, Standard deviation Confidence coefficient;

[0097] Multiply the upper limit of the predicted load rate by the rated heat exchange capacity of the heat exchange unit to convert it into the heat required on the primary side;

[0098] The expression is: ; The amount of heat required for the primary side. This refers to the rated heat exchange capacity of the unit.

[0099] The required primary flow rate for each time step is calculated based on the heat required on the primary side, and the required primary flow rate is mapped to the primary side valve opening. The mapping method is to obtain the corresponding opening through the valve characteristic curve. The valve characteristic curve is obtained through pre-calibration. Specifically, under the preset differential pressure condition, the corresponding flow rate value is collected according to different valve openings, and the correspondence between valve opening and flow rate value is established.

[0100] The formula for calculating the primary flow rate is: In the formula, For primary side flow, The amount of heat required for the primary side. To set the supply and return water temperature difference, For the density of water, This is the specific heat capacity of water.

[0101] The obtained primary valve opening is subject to valve opening limit and smoothing limit to generate the final primary valve opening.

[0102] The method for generating the final primary valve opening includes:

[0103] For each time step, the calculated valve opening is checked to see if it exceeds the range of the valve's minimum and maximum allowable opening. If the calculated value is lower than the valve's minimum allowable opening, the valve opening is set to the minimum allowable opening; if the calculated value is higher than the valve's maximum allowable opening, the valve opening is set to the maximum allowable opening.

[0104] The maximum variation range of the valve opening is preset. The difference between the valve opening at the current time step and the valve opening at the previous time step must not exceed the preset maximum variation range. If it exceeds the maximum variation range, a valve opening variation constraint interval is constructed at the current time step according to the preset maximum variation range. The primary valve opening is then corrected to the boundary value corresponding to the variation constraint interval to obtain the final primary valve opening at the current time step.

[0105] The heat required on the primary side is taken as the heat required on the secondary side. The secondary side circulating water flow rate is calculated using the pipeline heat balance formula. The calculated secondary side circulating water flow rate is mapped to the pump drive frequency using the linear proportional relationship of pump characteristics.

[0106] It should be noted that the linear proportional relationship of the pump characteristics refers to the flow-frequency mapping relationship established based on the approximate linear correspondence between the pump drive frequency and the output flow rate under stable operating conditions within the variable frequency operating range of the target circulating water pump. This mapping relationship can be obtained through pre-calibration, specifically: after the secondary circulating water pump is installed, the frequency converter is controlled to drive the circulating water pump at multiple preset frequency points. At each frequency point, the actual circulating water flow rate corresponding to the stable state is collected, and a correspondence is established between each frequency point and the corresponding flow rate value; within the variable frequency operating range, the collected frequency-flow rate data is linearly fitted to obtain the linear proportional relationship between the pump drive frequency and the circulating water flow rate.

[0107] The formula for calculating the secondary circulating water flow rate is: In the formula, For the required flow rate on the secondary side, For the heat that the secondary side pipeline needs to transport, For the temperature difference between the supply and return water on the secondary side, For the density of water, This is the specific heat capacity of water.

[0108] The method for obtaining the heat pump water supply temperature setpoint is as follows:

[0109] The acquired primary heat requirement, secondary circulating water flow rate, building thermal inertia parameters, and user behavior data are input into a pre-built water supply temperature prediction model, which outputs the heat pump water supply temperature setpoint for the current time step.

[0110] The training method for the water supply temperature prediction model includes:

[0111] Gradient boosting regression tree was selected as the water supply temperature prediction model to predict the heat pump water supply temperature setpoint for the current time step based on the heat required on the primary side, the circulating water flow on the secondary side, the building thermal inertia parameters, and user behavior data.

[0112] Before training the water supply temperature prediction, initialization is performed, and the initial hyperparameters are set as follows: number of decision trees 100-150, maximum depth of a single tree 4-6, minimum number of samples for node splitting 8-12, maximum number of features considered during splitting 3, learning rate 0.05-0.1, and regularization coefficient (L2) 0.1-0.2.

[0113] The mean squared error (MSE) loss function is used in the water supply temperature prediction training to measure the deviation between the model's predicted heat pump water supply temperature setpoint and the actual heat pump water supply temperature setpoint. Sufficient historical operational data is pre-collected as training data, including primary side heat requirement, secondary side circulating water flow rate, building thermal inertia parameters, user behavior data, and the actual heat pump water supply temperature setpoint used at the corresponding time step. The primary side heat requirement, secondary side circulating water flow rate, building thermal inertia parameters, and user behavior data are used as model inputs, with the actual heat pump water supply temperature setpoint used at the corresponding time step as the training target. The model is trained using the training set. Each new tree aims to fit the regression residual of the training set's loss function. The optimal splitting feature is selected using the MSE criterion to divide the samples into different child nodes until stopping conditions are met, including reaching the maximum depth and the number of samples in the child node being less than the minimum number of samples.

[0114] Gradient descent is used to optimize the weights of the leaf nodes of each new tree. The contribution of the new tree to the final prediction result is controlled by the learning rate, so as to avoid a single tree dominating the output and improve the stability of temperature prediction.

[0115] Bayesian optimization is used to search for the optimal combination of hyperparameters within a preset range. The optimization objective is to minimize the root mean square error of the validation set. The hyperparameter optimization range is as follows: number of decision trees 80-180, maximum depth of a single tree 3-7, learning rate 0.03-0.12, and regularization coefficient (L2) 0.05-0.25.

[0116] An early stopping mechanism is implemented during training, dividing all training data into training, validation, and test sets in a 7:2:1 ratio. The root mean square error (RMSE) of the validation set is calculated every 20 trees. Training stops when the RMSE decreases by less than 0.001 for three consecutive iterations to prevent overfitting. After training, the model parameters with the lowest RMSE are saved, including the splitting rules and leaf node weights of all decision trees, ensuring the model maintains stable and reliable water supply temperature prediction capabilities even for unseen heating operation conditions.

[0117] During the model evaluation phase, the trained model is validated using an independent test set, and the root mean square error, mean absolute error, and coefficient of determination are calculated. If the root mean square error of the test set is ≤3% of the target water supply temperature, the mean absolute error is ≤2% of the target water supply temperature, and the coefficient of determination is ≥0.95, then the model performance evaluation meets the standards and can be deployed for use, outputting the heat pump water supply temperature setpoint for the current time step based on the input parameters.

[0118] The calculated sequence of primary valve opening, secondary circulating water pump frequency, and heat pump supply temperature setpoint is sent to the controller.

[0119] To illustrate the generation process of the automatic control strategy, an example is given: At a certain current time step, the predicted average load rate is 0.71, the upper bound of the corresponding confidence interval is 0.74, and the rated heat exchange capacity of the heat exchange unit is 1000kW. Then, the heat required on the primary side at the current time step is 740kW.

[0120] Furthermore, with the supply and return water temperature difference set at 20℃ and the water density taken as 1000kg / m³, 3 When the specific heat capacity of water is taken as 4.2 kJ / (kg·℃), the required primary flow rate at the current time step can be calculated to be approximately 31.7 m³ / s. 3 / h; then, based on the valve characteristic curve, 31.7m 3 / h is mapped to a primary side valve opening of 56%. When the primary side valve opening meets the current allowable opening range and smooth change requirements, 56% can be determined as the final primary side valve opening at the current time step. Continuing to use 740kW as the required heat input for the secondary side, with a secondary side supply and return water temperature difference of 10℃, the secondary side circulating water flow rate can be calculated to be approximately 63.4m³. 3 / h; then, based on the linear proportional relationship of the pump characteristics, 63.4m 3 / h is mapped to the pump drive frequency of 37Hz.

[0121] Receive the terminal device adjustment command, integrate the terminal device adjustment command with the automatic control strategy, execute it according to the priority rules, and obtain the final execution opening of the terminal device at the current time step;

[0122] It should be noted that the end device can be a bridging valve.

[0123] The method for integrating the end-device adjustment command with the automatic control strategy and executing it according to priority rules to obtain the final execution opening degree of the end-device at the current time step includes:

[0124] The system receives user-sent terminal device adjustment commands via a remote APP platform. Each command includes a timestamp and a corresponding building unit identifier. The received commands are then converted into standardized control data. This standardized control data includes at least the terminal device number, target building unit identifier, command type, command activation time, command duration, and corresponding adjustment amount. Standardizing these terminal device adjustment commands allows for matching user-side input information with the controller's automatic control strategy within a unified data structure, providing a data foundation for subsequent integrated control.

[0125] Based on the current time step, local control parameters corresponding to the building unit where the target terminal device is located are extracted from the automatic control strategy already issued to the controller. These local control parameters include the primary-side valve opening, secondary-side circulating water pump frequency, heat pump supply temperature setpoint, and the target terminal device's basic opening at the current time step. The target terminal device's basic opening is an automatic control opening generated by the system without user intervention, based on the predicted load rate, building unit heat load distribution, secondary-side circulating water flow distribution, and historical temperature control response data for the area corresponding to the terminal device. It characterizes the target state the terminal device should be in to meet the predicted heating demand of the building unit. Preferably, the target terminal device's basic opening can be pre-calculated and cached in the controller or database, and directly invoked at the current time step.

[0126] The user-sent end-device adjustment command is mapped to the corresponding opening correction value. If the user sends an emergency manual opening command, the emergency manual opening command is directly used as the candidate control value for the end-device; wherein, the emergency manual opening command refers to the manual intervention command that the user directly inputs or selects a specific opening value for the target end-device in a remote APP platform, on-site human-machine interaction terminal or control management interface, for real-time adjustment of the opening of the end-device at the current moment.

[0127] If the user sends a mode switching command, the system calls the preset opening correction coefficient table corresponding to the mode to convert the mode command into an opening correction amount for the end device. It should be noted that the preset opening correction coefficient table corresponding to the mode is pre-established and stored in a database. Different modes correspond to different opening correction coefficients or opening ranges; for example, comfort mode corresponds to a larger opening correction amount, while energy-saving mode corresponds to a smaller opening correction amount. By uniformly mapping different types of user commands to end device opening correction amounts, it facilitates subsequent unified integration with automatic control strategies.

[0128] Establish a priority rule between the end device adjustment command and the automatic control strategy, and fuse the opening correction amount and the automatic control strategy according to the priority rule to obtain the final execution opening of the end device at the current time step.

[0129] The priority rules include a multi-level control logic: priority for emergency manual commands, secondary priority for mode switching commands, and priority for the basic automatic control strategy. Specifically, when an emergency manual opening command is received, it is prioritized to generate the final opening degree of the terminal device; when a mode switching command is received, it is used as a limited correction to the basic opening degree of the terminal device; when no user command is input, the terminal device executes according to the basic opening degree output by the automatic control strategy. By setting multi-level priority rules, both the personalized adjustment needs of users can be met, and unrestrained disruption of the system's automatic control logic by user commands can be prevented.

[0130] The primary valve opening fine-tuning amount is generated based on the final execution opening degree, and the updated control strategy is sent to the controller.

[0131] Methods for generating primary valve opening fine-tuning amounts based on the final execution opening include:

[0132] The final terminal device opening offset is obtained, and the terminal device opening offset is combined with the terminal device flow characteristic relationship to obtain the local branch flow offset. The primary valve opening fine adjustment is then recalculated. Specifically, the terminal device flow characteristic relationship is obtained through pre-calibration, which involves collecting the branch flow data corresponding to the terminal device under different opening conditions and establishing a correspondence table between the opening value and the branch flow value. When the flow offset needs to be calculated, the basic branch flow is first retrieved based on the basic opening value, and then the corrected branch flow is retrieved based on the corrected target opening value. The difference between the two is taken as the local branch flow offset.

[0133] It should be noted that the terminal device opening offset is: the final execution opening of the terminal device in the current time step minus the base opening of the target terminal device in the current time step; when the offset is positive, it means that the user requests an increase in the flow rate of this local area; when the offset is negative, it means that the user requests a decrease in the flow rate of this local area.

[0134] The local branch flow offset, the secondary side supply and return water temperature difference at the current time step, and the building thermal inertia parameters of the building unit corresponding to the terminal device are input into the pre-constructed offset prediction model, and the additional heat demand offset is output. The additional heat demand offset is used to characterize the change in heat demand caused by the local adjustment of the terminal device.

[0135] The training method for the offset prediction model includes: selecting a gradient boosting regression tree as the offset prediction model, which is used to predict the additional heat demand offset of the target area based on the local branch flow offset, the temperature difference between the secondary side supply and return water at the current time step, and the building thermal inertia parameters of the building unit corresponding to the terminal device.

[0136] Initialize the model before training by setting the following initial hyperparameters: number of decision trees 100-150, maximum depth of a single tree 4-6, minimum number of samples for node splitting 8-12, maximum number of features to consider during splitting 3, learning rate 0.05-0.1, and regularization coefficient (L2) 0.1-0.2.

[0137] The model training uses mean squared error as the loss function to measure the deviation between the predicted and actual additional heat demand offsets. Sufficient historical deviation data is pre-collected as training data, including local branch flow offsets, secondary side supply and return water temperature differences at the current time step, building thermal inertia parameters of the building unit corresponding to the terminal device, and the actual additional heat demand offset at the corresponding time step. The local branch flow offsets, secondary side supply and return water temperature differences, and building thermal inertia parameters are used as model inputs, while the actual additional heat demand offset at the corresponding time step is used as the training target. The model is trained using the training set. Each new tree aims to fit the regression residuals of the training set's loss function. The optimal splitting feature is selected using the mean squared error criterion to divide the samples into different child nodes until stopping conditions are met, including reaching the maximum depth and the number of samples in the child node being less than the minimum number of samples.

[0138] Gradient descent is used to optimize the weights of the leaf nodes of each new tree. The contribution of the new tree to the final prediction result is controlled by the learning rate, avoiding a single tree dominating the output and improving the stability of the offset prediction.

[0139] Bayesian optimization is used to search for the optimal combination of hyperparameters within a preset range. The optimization objective is to minimize the root mean square error of the validation set. The hyperparameter optimization range is as follows: number of decision trees 80-180, maximum depth of a single tree 3-7, learning rate 0.03-0.12, and regularization coefficient (L2) 0.05-0.25.

[0140] An early stopping mechanism is implemented during training, dividing all training data into training, validation, and test sets in a 7:2:1 ratio. The root mean square error (RMSE) of the validation set is calculated every 20 trees. Training stops when the RMSE decreases by less than 0.001 for three consecutive iterations to prevent overfitting. After training, the model parameters with the lowest RMSE on the validation set are saved, including the splitting rules and leaf node weights of all decision trees, ensuring the model maintains stable and reliable predictive capabilities for unseen thermal system operating conditions.

[0141] During the model evaluation phase, the trained model is validated using an independent test set, and the root mean square error, mean absolute error, and coefficient of determination are calculated. If the root mean square error of the test set is ≤3% of the target additional heat demand offset, the mean absolute error is ≤2% of the target additional heat demand offset, and the coefficient of determination is ≥0.95, then the model performance evaluation meets the standards and can be deployed for use, outputting the additional heat demand offset of the target area based on the input parameters.

[0142] A preset correction threshold is used to determine whether the offset of the additional heat demand exceeds the preset correction threshold. It should be noted that the correction threshold is set by those skilled in the art based on actual conditions or obtained through simulation of a large amount of data.

[0143] If the preset correction threshold is not exceeded, the current automatic control strategy remains unchanged, and only the local offset control of the terminal device is executed; if the preset correction threshold is exceeded, the heat compensation ratio is obtained according to the ratio of the additional heat demand offset to the current rated heat exchange capacity of the heat exchange unit, and then the heat compensation ratio is mapped to the primary side valve opening fine adjustment amount in combination with the primary side valve opening characteristic curve.

[0144] Specifically, if the additional heat demand offset is positive, the opening of the primary valve is increased; if the additional heat demand offset is negative, the opening of the primary valve is decreased.

[0145] The updated primary valve opening, secondary circulating water pump frequency, heat pump supply temperature setpoint, and final execution opening of the terminal device are all sent to the controller for execution.

[0146] In this embodiment, the terminal device opening offset is obtained based on the difference between the final execution opening degree of the target terminal device and its basic opening degree. Combined with a pre-established flow characteristic relationship of the terminal device, this terminal device opening offset is converted into a local branch flow offset. This further transforms the opening adjustment change at the terminal device level into a flow disturbance that characterizes the actual heat exchange and transport capacity change of the local branch. Compared to existing technologies that treat user-side terminal adjustment as a local, independent action, this invention does not stop at the terminal execution result itself but further identifies the substantial impact of the terminal adjustment on the heat transport state of the local branch. Therefore, it can provide a physically meaningful intermediate transmission quantity for subsequent global linkage control, avoiding the problem that local opening changes cannot be effectively transmitted to the unit-side control link.

[0147] In this invention, the offset of the final execution opening of the target terminal device relative to the basic opening is converted into a local branch flow offset. Combined with the secondary side supply and return water temperature difference and building thermal inertia parameters, the additional heat demand offset caused by local terminal adjustment is obtained, thus establishing an explicit transmission link between personalized terminal adjustment and the overall heating control of the unit side. Furthermore, by comparing the additional heat demand offset with a preset correction threshold, it is determined whether a local disturbance needs to trigger primary side linkage correction. This ensures that minor disturbances are absorbed only at the terminal, while larger disturbances are transmitted to the primary side valve opening for fine-tuning. This adjustment is then executed in conjunction with the secondary side circulating water pump's basic frequency, the heat pump supply water temperature setpoint, and the final execution opening of the terminal device. This improves the coordination between local adjustment and overall heating control, reduces system control fluctuations, and enhances heating stability and the rationality of heat distribution.

[0148] Furthermore, this invention does not treat user-side terminal device adjustment as an isolated operation separate from automatic control. Instead, it integrates terminal device adjustment commands with automatic control strategies according to priority rules, and performs coordinated fine-tuning of primary-side valve openings based on the final execution opening degree of the terminal device, local branch flow offsets, and additional heat demand offsets. This allows personalized adjustment needs in local areas to be incorporated into the overall unit control chain, ensuring consistency between local manual intervention and automatic control based on load rate prediction, and preventing disorderly impacts on the overall heat distribution of the system caused by local adjustments. This technical solution improves both the personalized heating control capability of the target area and enhances the coordination and operational stability of the entire heating system under complex conditions, thereby achieving a balance between improved heating quality and optimized system operating efficiency.

[0149] Example 2

[0150] Please see Figure 4This invention provides a technical solution: a heat pump-type secondary side water supply temperature closed-loop regulation system, which is used to implement the heat pump-type secondary side water supply temperature closed-loop regulation method, including:

[0151] The historical data acquisition module acquires historical load characteristic data of the target area, including historical meteorological data, user behavior data, building thermal inertia parameters, and historical heat exchanger unit load rates.

[0152] The model training module analyzes the acquired historical load characteristic data, trains a recurrent neural network model, and predicts the load rate prediction sequence of the heat exchanger unit at time k in the future.

[0153] The load rate prediction module, based on a trained recurrent neural network model, obtains the predicted load rate curve and the range of uncertainty.

[0154] The strategy generation module generates automatic control strategies for primary valve opening, secondary circulating water pump frequency, and heat pump supply temperature setpoints based on the predicted load rate curve and uncertainty range, and sends them to the controller.

[0155] The data fusion module receives the adjustment instructions from the terminal device, merges the terminal device adjustment instructions with the automatic control strategy, executes them according to priority rules, and obtains the final execution opening of the terminal device at the current time step.

[0156] The fine-tuning execution module generates a fine-tuning amount for the primary valve opening based on the final execution opening. It then sends the updated primary valve opening, secondary circulating water pump frequency, heat pump supply temperature setpoint, and final execution opening of the terminal device to the controller for execution.

[0157] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended scope and its equivalents.

Claims

1. A heat pump-type closed-loop regulation method for secondary-side water supply temperature, applied to a central hot water heating system including a heat pump unit, a plate heat exchanger unit, a primary-side circulation loop, a secondary-side circulation loop, and terminal devices, characterized in that... include: Acquire historical load characteristic data for the target area, including historical meteorological data, user behavior data, building thermal inertia parameters, and historical heat exchanger unit load rates; The acquired historical load characteristic data is analyzed, a recurrent neural network model is trained, and the load rate prediction sequence of the heat exchanger unit at time k is predicted. Based on the trained recurrent neural network model, the predicted load rate curve and uncertainty range are obtained; An automatic control strategy is generated based on the predicted load rate curve and the uncertainty range to control the primary valve opening, secondary circulating water pump frequency and heat pump supply temperature setpoint, and then sent to the controller. Receive the terminal device adjustment command, integrate the terminal device adjustment command with the automatic control strategy, execute it according to the priority rules, and obtain the final execution opening of the terminal device at the current time step; Based on the final execution opening, the primary valve opening fine-tuning amount is generated, and the updated primary valve opening, secondary circulating water pump frequency, heat pump supply water temperature setpoint, and final execution opening of the terminal device are jointly sent to the controller for execution. The method for generating the primary valve opening fine-tuning amount based on the final execution opening includes: Obtain the final terminal device opening offset and convert it to obtain the local branch flow offset; The end-point device opening offset is: the final execution opening of the end-point device at the current time step minus the base opening of the target end-point device at the current time step; The local branch flow offset, the secondary side supply and return water temperature difference at the current time step, and the building thermal inertia parameters of the building unit corresponding to the terminal device are input into the pre-built offset prediction model, and the additional heat demand offset is output. If the offset of additional heat demand does not exceed the preset correction threshold, the current automatic control strategy will remain unchanged, and only the local offset control of the terminal device will be executed. If the preset correction threshold is exceeded, the heat compensation ratio is obtained based on the ratio of the additional heat demand offset to the current rated heat exchange capacity of the heat exchange unit. Combined with the primary side valve opening characteristic curve, the heat compensation ratio is mapped to the primary side valve opening fine adjustment amount.

2. The closed-loop regulation method for secondary side water supply temperature of a heat pump according to claim 1, characterized in that, Methods for predicting the load rate of heat exchanger units at time k include: Based on the historical load characteristic data, a training sample set is constructed using a sliding window approach; Using multi-source feature vectors from T consecutive historical moments as input samples, labeled as the heat exchanger unit load rate sequence at future k moments, with a sliding window step size of s hours, multiple training samples are generated sequentially, and the training samples are divided into training set, validation set and test set. The training samples are input into a recurrent neural network model for training. The recurrent neural network model includes a recurrent layer containing hidden units, and outputs a load rate prediction sequence for the future k time step through a fully connected layer.

3. The closed-loop regulation method for secondary side water supply temperature of a heat pump according to claim 1, characterized in that, The method for obtaining the predicted load factor curve includes: using a multi-source feature vector sequence of T consecutive historical moments as the input sequence of a recurrent neural network model, inputting the input sequence into the trained recurrent neural network model, and obtaining the load factor prediction sequence for the next k time steps; The input sequence is introduced with random perturbation and Monte Carlo sampling to generate multiple sets of perturbation input sequences. Each set of perturbation input sequences is then input into the recurrent neural network model to obtain multiple sets of load rate prediction results corresponding to the next k time steps. The average load rate of multiple load rate prediction results corresponding to each time step is calculated and used as the predicted load rate point value for that time step. The average load rate of each time step is connected to form a load rate prediction curve.

4. The closed-loop regulation method for secondary side water supply temperature of a heat pump according to claim 3, characterized in that, The methods for obtaining the uncertainty range include: The standard deviation is calculated for the predicted sample sequence at each time step to quantify the fluctuation range of the predicted value at that time step. Based on the standard deviation and combined with the preset confidence coefficient, the upper and lower bounds of the confidence interval for each time step are calculated, and the corresponding upper and lower confidence intervals are output as the uncertainty range.

5. The closed-loop regulation method for secondary side water supply temperature of a heat pump according to claim 1, characterized in that, The method for obtaining the primary valve opening and the secondary circulating water pump frequency is as follows: Obtain the mean load rate and the upper bound of the confidence interval for the current time step, i.e., the upper limit of the predicted load rate; Multiply the upper limit of the predicted load rate by the rated heat exchange capacity of the heat exchange unit to convert it into the heat required on the primary side; Calculate the required primary flow rate at each time step based on the heat required on the primary side, and map the required primary flow rate to the primary valve opening. The obtained primary valve opening is subject to valve opening limit and smoothing limit to generate the final primary valve opening; The heat required on the primary side is taken as the heat required on the secondary side. The secondary side circulating water flow rate is calculated using the pipeline heat balance formula. The calculated secondary side circulating water flow rate is mapped to the pump drive frequency using the linear proportional relationship of pump characteristics.

6. The closed-loop regulation method for secondary side water supply temperature of a heat pump according to claim 5, characterized in that, The method for generating the final primary valve opening includes: For each time step, the calculated valve opening is checked to see if it exceeds the range of the valve's minimum and maximum allowable opening. If the calculated value is lower than the valve's minimum allowable opening, the valve opening is set to the minimum allowable opening; if the calculated value is higher than the valve's maximum allowable opening, the valve opening is set to the maximum allowable opening. The maximum variation range of the valve opening is preset. The difference between the valve opening at the current time step and the valve opening at the previous time step must not exceed the preset maximum variation range. If it exceeds the maximum variation range, a valve opening variation constraint interval is constructed at the current time step according to the preset maximum variation range. The primary valve opening is then corrected to the boundary value corresponding to the variation constraint interval to obtain the final primary valve opening at the current time step.

7. The closed-loop regulation method for secondary side water supply temperature of a heat pump according to claim 5 or 6, characterized in that, The method for obtaining the heat pump water supply temperature setpoint is as follows: The acquired primary heat requirement, secondary circulating water flow rate, building thermal inertia parameters, and user behavior data are input into a pre-built water supply temperature prediction model, which outputs the heat pump water supply temperature setpoint for the current time step.

8. The closed-loop regulation method for secondary side water supply temperature of a heat pump according to claim 1, characterized in that, The method for integrating the end-device adjustment command with the automatic control strategy and executing it according to priority rules to obtain the final execution opening degree of the end-device at the current time step includes: Receive terminal device adjustment commands sent by users through a remote APP platform; Based on the current time step, extract the local control parameters corresponding to the building unit where the target terminal device is located from the automatic control strategy that has been issued to the controller; Map the user-sent end-device adjustment commands to the corresponding opening correction values; Establish a priority rule between the end device adjustment command and the automatic control strategy, and fuse the opening correction amount and the automatic control strategy according to the priority rule to obtain the final execution opening of the end device at the current time step.

9. The closed-loop regulation method for secondary side water supply temperature of a heat pump according to claim 8, characterized in that, The local control parameters include: the opening degree of the primary valve, the frequency of the secondary circulating water pump, the set value of the heat pump water supply temperature, and the basic opening degree of the target terminal device at the current time step.

10. A heat pump-type secondary side water supply temperature closed-loop regulation system, used to implement the heat pump-type secondary side water supply temperature closed-loop regulation method according to any one of claims 1-9, characterized in that, include: Historical data acquisition module: used to acquire historical load characteristic data of the target area, including historical meteorological data, user behavior data, building thermal inertia parameters and historical heat exchanger unit load rate; Model training module: used to analyze the acquired historical load characteristic data, train the recurrent neural network model, and predict the load rate prediction sequence of the heat exchanger unit at time k in the future; Load rate prediction module: used to obtain the predicted load rate curve and uncertainty range based on the trained recurrent neural network model; Strategy generation module: Used to generate automatic control strategies for primary valve opening, secondary circulating water pump frequency and heat pump supply temperature setpoints based on the predicted load rate curve and uncertainty range, and send them to the controller; Data fusion module: Used to receive adjustment instructions from the end device, fuse the adjustment instructions with the automatic control strategy, execute according to priority rules, and obtain the final execution opening degree of the end device at the current time step; Fine-tuning execution module: Used to generate a fine-tuning amount for the primary valve opening based on the final execution opening, and send the updated primary valve opening, secondary circulating water pump frequency, heat pump supply water temperature setpoint, and final execution opening of the terminal device to the controller for execution.

Images