A control method of a dual-source heat pump system

Abstract

The application belongs to the technical field of double-source heat pump, and provides a control method of a double-source heat pump system. Various sensors are arranged on field equipment, collected data is transmitted to an edge gateway for classification and mapping, preprocessing is performed by aligning time stamps, filling in missing items and smoothing high-frequency fluctuations; the collected data is spliced into a feature vector, the error of linear regression is calculated, the candidate values of hyperparameters are selected, a loss function is defined in combination with the error, the corresponding performance coefficients of different heat sources are obtained, the consumption cost, the temperature difference of the heat source and the performance coefficient are taken as score calculation items, and a decision of switching the heat source is generated according to the score; the decision is sent to a field execution unit, smooth switching is performed through dynamic load compensation and preheating preparation, the previous heat source is closed after the switching is completed, a decision log is exported at a fixed time to calculate the performance coefficient error, the error conditions of different heat sources are analyzed, a warning threshold is set, and the weight of the score calculation item is adjusted according to the warning information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dual-source heat pump, in particular to a control method of dual-source heat pump system. BACKGROUND

[0002] Dual-source heat pump is a system that combines the advantages of two different heat sources, which can intelligently switch or parallel two heat sources according to external environmental conditions, improve system performance, stability and operating efficiency.

[0003] In Chinese application No. CN201910782526.2, a dual-heat-source air conditioner heat pump system and its control method are disclosed. The configuration method includes a compressor, a four-way valve, a first heat exchanger, an outdoor heat exchanger, a first electronic expansion valve, a second heat exchanger, a second electronic expansion valve, a solar heat collector, and a heat storage tank. By changing the system flow and control strategy, the dual-heat-source utilization mode of the heat pump system is realized, the resources of solar energy and air source are reasonably utilized, and the heat pump power consumption is minimized. The heat recovery function in refrigeration mode is designed to preheat domestic water and provide comprehensive operating energy efficiency of the heat pump system.

[0004] In the field of dual-source heat pump technology, although there are technologies that change the system flow and control strategy to realize dual-heat-source heat pump system, reasonably utilize solar energy and air source resources, and reduce heat pump power consumption, in the context of building air conditioning and heating systems, traditional heat source systems are easily affected by temperature, geology and other factors during seasonal transition, and the efficiency difference between different seasons is large, which can easily lead to significant temperature fluctuations. At the same time, the existing dual-source heat pump system fails to provide an intelligent, reasonable and accurate method to schedule when switching heat sources, resulting in repeated oscillation between the two sources, and the suitable heat source energy efficiency is not fully utilized, causing energy waste. Therefore, there is a need for a control method that can monitor environmental temperature and load conditions in real time, intelligently judge and switch different heat source working modes to ensure high energy efficiency utilization in different seasons. SUMMARY

[0005] The present application provides a control method of dual-source heat pump system, which aims to solve the problem of single heat source efficiency being greatly affected by external factors and the lack of intelligent scheduling method when switching dual-source system.

[0006] The technical solution adopted by the present application to solve the above technical problems is to provide a control method of dual-source heat pump system, which includes:

[0007] Deploying sensors to collect data in real time and transmitting it to the cloud for preprocessing involves deploying various sensors on field devices, transmitting the collected data to the edge gateway for classification and mapping, assigning identifiers to each data set by floor and unit, and preprocessing the data by aligning the timestamps of each device, filling in missing data, and setting a sliding window to smooth high-frequency fluctuations.

[0008] The process of training the model to obtain performance coefficients and designing a scoring mechanism to generate decisions involves concatenating the collected data into a feature vector, normalizing it, inputting it into the model, calculating the error of linear regression, selecting candidate values ​​for hyperparameters, defining a loss function based on the error, obtaining corresponding performance coefficients for different heat sources, and using consumption costs, heat source temperature differences, and performance coefficients as scoring items. Based on the scores, decisions on switching heat sources are generated.

[0009] The process involves issuing decisions for smooth switching, collecting feedback periodically for iterative optimization, issuing decisions to field execution units, and using dynamic load compensation and preheating preparation for smooth switching. Once the switch is complete, the previous heat source is shut down. Decision logs are exported periodically to calculate performance coefficient errors, analyze errors from different heat sources, set early warning thresholds, and adjust the weights of scoring items based on early warning information.

[0010] As a preferred implementation, the specific steps for deploying sensors to collect data in real time are as follows: An air temperature sensor is deployed on the outdoor unit to collect ambient temperature; a current transformer is deployed in the compressor's power circuit to collect power in real time; and a flow sensor is deployed in the terminal fan coil unit piping to collect circulating flow. The collected sensor data is transmitted to the industrial edge gateway. By setting a unified data type, a unique group number is assigned to each data stream, and the data is classified by floor and unit, mapped to objects recognizable by the BMS. The BMS objects are aligned with timestamps using the NTP protocol. The sensor sends a request message with a local transmission timestamp to the edge gateway. After receiving the message, the edge gateway records the reception time and sends it back along with its own transmission time. After receiving the message, the sensor records its local reception time, calculates the delay and time difference based on the timestamp, and adjusts its local clock.

[0011] As a preferred implementation, the specific steps for transmitting the data to the cloud for preprocessing are as follows: The collected data after classification and mapping is reported to the cloud message queue in JSON format. Upon receiving a new message, the cloud first performs data verification, compares the validity of the timestamp, determines the range of the collected data values, converts the original values ​​into physical quantities, and uses linear interpolation to fill in missing values ​​in the collected data. The missing values ​​are filled in by averaging the time and value differences between two data points before and after the missing item, using the following formula:

[0012]

[0013] Where Dmissing D represents missing data points. before D represents the previous valid data point before the missing data point. after t represents the next valid data point after a missing data point. missing t represents the timestamp corresponding to the missing data point. after t represents the timestamp of the next valid data point after the missing data point. before Indicates the timestamp of the previous valid data point preceding the missing data point;

[0014] For high-frequency fluctuating data segments, a fixed-length sliding window is used to calculate the arithmetic mean of the sampled values ​​within each window to smooth out short-term jitter. The formula is:

[0015]

[0016] in Let M represent the smoothed value within the window at time w, M represent the length of the window, i represent the index of the i-th sampling point shifted forward from the current time w, and x represent the smoothed value within the window. w-i This is the original high-frequency data.

[0017] As a preferred implementation, the specific steps for obtaining performance coefficients in the training model are as follows: Pre-determine the sensor data units and order of the feature list; concatenate the pre-processed sensor data according to the feature list; iterate through the maximum and minimum values ​​of each sensor data point, mapping the data to values ​​from 0 to 1 using the maximum and minimum values ​​as intervals; construct the input vector from the normalized feature vector; divide the input vector into training, validation, and test sets; select a multiple linear regression model for training; characterize the linear relationship between the dependent variable and multiple independent variables, and fit the influence of different independent variables on the dependent variable; enumerate and select values ​​for the hyperparameters in the model; form a network by performing a Cartesian product of the selected candidate values ​​for evaluation; calculate the root mean square error between the predicted value of the multiple linear regression model and the actual data of the validation set, using the following formula:

[0018]

[0019] error squared Let N represent the root mean square error, N represent the number of samples selected, and j represent the j-th sample. Let y represent the predicted value of the j-th sample. i This represents the true value of the j-th sample;

[0020] Candidate values ​​for hyperparameters are selected based on the root mean square error. The model is trained using the selected hyperparameters and the training set. The absolute error between the predicted and actual values ​​is defined as the loss function. The loss value for each sample in the training set is calculated, and the average value is taken as the training error. The collected data in the training set is constructed into a matrix. The regression coefficient vector is obtained using the least squares method. A set of undetermined regression coefficient vectors is set. The regression coefficient vectors and the constructed matrix of the collected data are used for matrix calculation. When the training error is set to 0, the partial derivatives of the regression coefficient vectors are calculated. The gradient of the partial derivatives of the regression coefficient vectors is set to 0 and solved.

[0021] As a preferred implementation, the specific steps of the design scoring mechanism for generating decisions are as follows: For both ground-source and air-source heat sources, a multiple linear regression model is used to obtain the predicted COP; the energy consumption and mechanical wear caused by a single switch are converted into consumption costs; and a comprehensive score function is calculated for the two heat sources.

[0022] score a =α1COP a -α2cost switch -α3ΔT a , a∈{air,ground},

[0023] Among them, score a This represents the overall score of the a-th heat source, where α1 represents the performance coefficient weight, α2 represents the cost weight, α3 represents the temperature difference weight, and COP is... a Let cost represent the COP of the a-th heat source. switch ΔT represents the cost of switching heat sources. a The heat exchanger temperature difference of the a-th heat source is represented by air, and ground represents ground source.

[0024] A switching interval is set, and the current scores of different heat sources are calculated in real time. First, the switching interval is judged. If the time since the last switching is less than the set interval, no switching operation is performed. If the set interval is exceeded, a decision to switch heat sources is generated based on the scores. air >score ground When switching to the air source, when the score air <score ground When switching to ground source, when score air =score ground At that time, leave the current heat source unchanged.

[0025] As a preferred implementation, the specific steps for smooth switching by issuing decisions are as follows: After the edge gateway obtains the decision, it issues the decision to the field execution unit for preheating preparation. First, the temperature difference between the two sources is calculated, and the heat source to be switched is heated or cooled. When the temperature difference between the two sources is close, the valve ratio is adjusted to perform dynamic load compensation. The temperature difference value is collected, and the compensation amount is adjusted according to the difference value, increasing or decreasing the pump speed and fan speed. Subsequent difference values ​​are collected periodically to continue adjustment. After the valve adjustment and dynamic load compensation are completed, the switching status is determined, the valve of the previous heat source is closed, and the pipeline of the previous heat source is disconnected.

[0026] As a preferred implementation, the specific steps for periodically collecting feedback and iterative optimization are as follows: periodically exporting decision logs from the BMS, including COP predicted and actual values; switching the heat source type for decision-making; and calculating the COP error using the following formula:

[0027]

[0028] in This represents the performance coefficient error of the k-th data point. COP represents the prediction performance coefficient for the k-th data point. k This represents the actual performance coefficient of the k-th data item;

[0029] The mean and variance of COP error are calculated, classified by ground source and air source, and the deviation of different heat sources is analyzed. Warning thresholds are set for the mean and variance, and automatic alarms are issued for indicators that exceed the thresholds. The assessment results and indicators that exceed the thresholds are sent to the operation and maintenance department via SMS. The weights in the comprehensive scoring function are fine-tuned based on the warning information, error data is collected, and the data is input into the multiple linear regression model for retraining, thereby updating the parameters.

[0030] The beneficial effects of this invention are:

[0031] 1. By using the dual-source COP prediction model and dynamic scoring mechanism obtained through training, the heat source with priority scheduling performance can be implemented, thereby improving the overall system performance indicators and reducing power consumption.

[0032] 2. By introducing early warning thresholds and switching costs, frequent switching caused by minor COP fluctuations is avoided, and the pump speed is gradually increased and the valve position is adjusted in stages to improve the lifespan of key components.

[0033] Legend

[0034] Figure 1 This is a flowchart of a control method for a dual-source heat pump system. Detailed Implementation

[0035] To make the technical means, creative features, objectives, and effects of this invention easier to understand, the invention is further described below with reference to specific embodiments. However, the following embodiments are merely preferred embodiments of this invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments described herein without creative effort are all within the protection scope of this invention.

[0036] Example 1, as Figure 1 This is a control method for a dual-source heat pump system, including the following steps:

[0037] Deploy sensors to collect data in real time; transmit the data to the cloud for preprocessing.

[0038] Train the model to obtain performance coefficients; design a scoring mechanism to generate decisions;

[0039] Decisions are issued and transitions are made smoothly; feedback is collected regularly for iterative optimization.

[0040] The following are the specific implementation steps: A control method for a dual-source heat pump system, wherein the specific steps for deploying sensors to collect data in real time are as follows:

[0041] An air temperature sensor is deployed on the outdoor unit to collect ambient temperature data. A current transformer is deployed in the compressor's power circuit to collect power data in real time. A flow sensor is deployed in the terminal fan coil unit piping to collect circulating flow. The collected data from each sensor is transmitted to the industrial edge gateway. The edge gateway assigns a unique group number to each data source by setting a unified data type, classifies the data by floor and unit, and maps it into an object that can be recognized by the Building Management System (BMS). The BMS object is aligned with the timestamp using the Network Time Protocol (NTP). The sensor sends a request message with a local transmission timestamp to the edge gateway. After receiving the message, the edge gateway records the reception time and sends it back along with its own transmission time. After receiving the message, the sensor records the local reception time, calculates the delay and time difference based on the timestamp, and adjusts its local clock accordingly.

[0042] The specific steps for preprocessing the data transmitted to the cloud are as follows: The collected data, after classification and mapping, is reported to the cloud message queue in JSON format. Upon receiving a new message, the cloud first performs data verification, comparing the validity of the timestamp, determining the range of the collected data values, converting the original values ​​into physical quantities, and imputing missing values ​​in the collected data using linear interpolation. This is done by averaging the time and value differences between the two nearest valid data points before and after the missing item, using the following formula:

[0043]

[0044] Where D missing D represents missing data points. before D represents the previous valid data point before the missing data point. after t represents the next valid data point after a missing data point. missing t represents the timestamp corresponding to the missing data point. after t represents the timestamp of the next valid data point after the missing data point. before Indicates the timestamp of the previous valid data point preceding the missing data point;

[0045] For high-frequency fluctuating data segments, a fixed-length sliding window, such as the size of 5 sampling data points, is used. The arithmetic mean of the sampled values ​​within each window is calculated to smooth out short-term jitter. The formula is:

[0046]

[0047] in Let M represent the smoothed value within the window at time w, M represent the length of the window, i represent the index of the i-th sampling point shifted forward from the current time w, and x represent the smoothed value within the window. w-i This is the original high-frequency data.

[0048] Based on the above steps, the specific steps for training the model and obtaining its performance coefficients are as follows:

[0049] The sensor data units and order of the feature list are predefined. Preprocessed sensor data are concatenated according to the feature list. For each sensor data point, the maximum and minimum values ​​are iterated, and the data are mapped to values ​​from 0 to 1 using the maximum and minimum values. The normalized feature vectors are used to construct the input vector. The input vector is divided into training, validation, and test sets in an 8:2:2 ratio. A multiple linear regression model is selected for training. By characterizing the linear relationship between the dependent variable and multiple independent variables, and fitting the influence of different independent variables on the dependent variable, real-time prediction is achieved. Hyperparameters such as learning rate, maximum depth, number of blade points, and column sampling ratio are enumerated and selected. The selected candidate values ​​are then used to form a Cartesian product to form a network for evaluation. The root mean square error between the predicted value of the multiple linear regression model and the actual data of the validation set is calculated using the following formula:

[0050]

[0051] error squared Let N represent the root mean square error, N represent the number of samples selected, and j represent the j-th sample. Let y represent the predicted value of the j-th sample. i This represents the true value of the j-th sample;

[0052] The candidate values ​​with the smallest root mean square error are selected as the optimal hyperparameters. The model is trained using the optimal hyperparameters and the training set. The absolute error between the predicted value and the actual value is defined as the loss function. The loss value of each sample in the training set is calculated, and the average value is taken as the training error. The collected data in the training set is constructed into a matrix. The least squares method is used to solve for the predicted performance coefficient (COP). Specifically, a set of regression coefficient vectors to be determined is set. The regression coefficient vectors and the constructed matrix of the collected data are used to perform matrix calculation. When the training error is set to 0, the partial derivative of the regression coefficient vector is calculated, and its gradient is set to 0 to obtain the equation. The solution obtained is the performance coefficient.

[0053] The specific steps for the scoring mechanism to generate decisions are as follows: For both ground-source and air-source heat sources, a multiple linear regression model is used to obtain the predicted COP; the energy consumption and mechanical wear caused by a single switch are converted into consumption costs; and a comprehensive scoring function is calculated for both heat sources.

[0054] score a =α1COP a -α2cost switch -α3ΔT a , a∈{air,ground},

[0055] Among them, score a This represents the overall score of the a-th heat source, where α1 represents the performance coefficient weight, α2 represents the cost weight, α3 represents the temperature difference weight, and COP is... a Let cost represent the COP of the a-th heat source. switch ΔT represents the cost of switching heat sources. a The heat exchanger temperature difference of the a-th heat source is represented by air, and ground represents ground source.

[0056] Set a switching interval, such as 20 minutes, and calculate the current scores of different heat sources in real time. First, determine the switching interval: if the time since the last switching is less than the set interval, no switching operation is performed; if the time since the last switching exceeds the set interval, a decision to switch heat sources is generated based on the scores. air >score ground When switching to the air source, when the score air <score ground When switching to ground source, when score air =score ground At that time, leave the current heat source unchanged.

[0057] Based on the above steps, the specific steps for issuing the decision to smoothly switch over are as follows:

[0058] After the edge gateway obtains the decision, it sends the decision to the field execution unit, such as the control valve driver and the control frequency converter, to carry out preheating preparation. First, the temperature difference between the two sources is calculated, and the heat source to be switched is heated or cooled. When the temperature difference between the two sources is close, the valve ratio is adjusted to perform dynamic load compensation. Specifically, the temperature difference value is collected, and the compensation amount is adjusted according to the difference value, increasing or decreasing the pump speed and fan speed. Subsequent difference values ​​are collected periodically to continue the adjustment. After the valve adjustment and dynamic load compensation are completed, the switching status is determined, the valve of the previous heat source is closed, and the pipeline of the previous heat source is disconnected.

[0059] The specific steps for regularly collecting feedback and iterative optimization are as follows: Export the decision log from the BMS at 00:00 every day, including the predicted and actual COP values, changes in heat source types, etc., and calculate the COP error using the following formula:

[0060]

[0061] in This represents the performance coefficient error of the k-th data point. COP represents the prediction performance coefficient for the k-th data point. k This represents the actual performance coefficient of the k-th data item;

[0062] The mean and variance of COP error are calculated, categorized by ground and air sources, and the deviation of different heat sources is analyzed. Warning thresholds are set for the mean and variance, respectively. For example, the variance threshold for air heat sources can be 0.16, which is the standard deviation of the calculated data. A factor of 1.3 is set as the amplification factor to counteract prediction noise, and the result is obtained by multiplying it by the standard deviation. Automatic alarms are issued for indicators exceeding the thresholds, and the assessment status and indicators exceeding the thresholds are sent to the operation and maintenance department via SMS. The weights in the comprehensive scoring function are fine-tuned based on the alarm frequency and the deviation from the threshold. Error data is collected and merged with historical data to form a new training set. Unified preprocessing and feature construction are performed on the new training set, and it is input into the multiple linear regression model for retraining, thereby updating the parameters.

[0063] Example 2, based on Example 1 above, describes the application of a control method for a dual-source heat pump system in a building air conditioning and heating system scenario, specifically as follows:

[0064] Step one involves installing sensors on the central air conditioning unit, air handling unit, and terminal fan coil units to collect data such as temperature, flow rate, and current. These sensors are connected to the field gateway device via the RS-485 industrial protocol. The RS-485 protocol specifies the electrical characteristics of the bus interface. Every 30 seconds, the gateway device synchronizes the collected data with a Coordinated Universal Time (UTC) timestamp and pushes it to the cloud for processing. In the cloud, the collected data is processed in predefined units such as °C and m³. 3 The values ​​ / h, Pa, and A are converted, and the resulting numerical values ​​are then interpolated and smoothed.

[0065] Step two involves aggregating and splicing the pre-processed data collected in the cloud to construct a feature vector containing parameters such as ambient temperature, supply air temperature, compressor power, and chilled water flow rate. An ensemble regression tree is then used to minimize the training error. Specifically, an ensemble model is established, representing the overall prediction function as a weighted sum of multiple regression trees. The squared loss is defined as the target of the training error. The residual values ​​of the models at different stages are calculated, and the optimal parameters of the ensemble regression tree model are minimized. The product of the predicted noise standard deviation ε and the noise coefficient μ is used as the switching threshold. A switching decision is made only when the difference between the two heat source scores is greater than the switching threshold; otherwise, no action is taken. air -score ground When the value is greater than ε·μ, switch to the air source. When the score... ground -score air When the value is greater than ε·μ, switch to ground source.

[0066] Step 3: Based on the scoring mechanism, a decision is generated and sent from the BMS to the field equipment. During the heat source switching process, taking the switch from air source to ground source as an example, when the air source is running, the valve of the ground source heat exchanger is gradually opened, and water at the current temperature is slowly injected into the ground source until the ground source temperature is close to the current air source temperature. At the same time, the ground source circulation pump speed is gradually increased. When the valve position is fully at the ground edge, the switch is considered complete. If the system has no continuous oscillation or the temperature and flow rate do not meet the standards within 10 minutes, the switch is considered to have failed, and an early warning message is generated and reported to the cloud. The BMS sends the collected data such as the ratio of predicted COP to measured thermal efficiency, the number of switches, and the success rate to the cloud for calculation. A composite loss function is constructed, which includes a prediction error term, a switching frequency penalty term, and a switching success reward term. The corresponding loss value is calculated. Each iteration uses the model parameters corresponding to the minimum loss value as the optimal parameters, and the iteration is repeated until the loss value converges, that is, approaches 0.

[0067] The embodiments of the present invention described above are subject to modification and change of method by those skilled in the art without departing from the embodiments and broader aspects of the present invention. The appended claims are intended to include all such modifications and changes of method that do not depart from the present invention.

Claims

1. A control method for a dual-source heat pump system, characterized in that, Includes the following steps: Deploying sensors to collect data in real time and transmitting it to the cloud for preprocessing involves deploying various sensors on field devices, transmitting the collected data to the edge gateway for classification and mapping, assigning identifiers to each data set by floor and unit, and preprocessing the data by aligning the timestamps of each device, filling in missing data, and setting a sliding window to smooth high-frequency fluctuations. The process of training the model to obtain performance coefficients and designing a scoring mechanism to generate decisions involves concatenating the collected data into a feature vector, normalizing it, inputting it into the model, calculating the error of linear regression, selecting candidate values ​​for hyperparameters, defining a loss function based on the error, obtaining corresponding performance coefficients for different heat sources, and using consumption costs, heat source temperature differences, and performance coefficients as scoring items. Based on the scores, decisions on switching heat sources are generated. The process involves issuing decisions for smooth switching, collecting feedback periodically for iterative optimization, issuing decisions to field execution units, and using dynamic load compensation and preheating preparation for smooth switching. Once the switch is complete, the previous heat source is shut down. Decision logs are exported periodically to calculate performance coefficient errors, analyze errors from different heat sources, set early warning thresholds, and adjust the weights of scoring items based on early warning information.

2. The control method for a dual-source heat pump system according to claim 1, characterized in that: The specific steps for deploying sensors to collect data in real time are as follows: An air temperature sensor is deployed on the outdoor unit to collect ambient temperature data. A current transformer is deployed in the compressor's power circuit to collect power data in real time. A flow sensor is deployed in the terminal fan coil unit piping to collect circulating flow data. The collected data from each sensor is transmitted to the industrial edge gateway. By setting a unified data type, a unique group number is assigned to each data stream, and the data is classified by floor and unit, mapped into objects that the BMS can recognize. The BMS objects are aligned with timestamps using the NTP protocol. The sensor sends a request message with a local transmission timestamp to the edge gateway. After receiving the message, the edge gateway records the reception time and sends it back along with its own transmission time. After receiving the message, the sensor records its local reception time, calculates the delay and time difference based on the timestamp, and adjusts its local clock accordingly.

3. The control method for a dual-source heat pump system according to claim 1, characterized in that: The specific steps for transmitting data to the cloud for preprocessing are as follows: The collected data, after classification and mapping, is reported to the cloud message queue in JSON format. Upon receiving a new message, the cloud first performs data verification, comparing the validity of the timestamp, determining the range of the collected data values, converting the original values ​​into physical quantities, and imputing missing values ​​in the collected data using linear interpolation. This is done by averaging the time and value differences between two data points before and after a missing item, using the following formula: Where D missing D represents missing data points. before D represents the previous valid data point before the missing data point. after t represents the next valid data point after a missing data point. missing t represents the timestamp corresponding to the missing data point. after t represents the timestamp of the next valid data point after the missing data point. before This represents the timestamp of the previous valid data point before the missing data point.

4. The control method for a dual-source heat pump system according to claim 3, characterized in that: The specific steps for transmitting data to the cloud for preprocessing also include: For high-frequency fluctuating data segments, a fixed-length sliding window is used to calculate the arithmetic mean of the sampled values ​​within each window to smooth out short-term jitter. The formula is as follows: in Let M represent the smoothed value within the window at time w, M represent the length of the window, i represent the index of the i-th sampling point shifted forward from the current time w, and x represent the smoothed value within the window. w-i This is the original high-frequency data.

5. The control method for a dual-source heat pump system according to claim 1, characterized in that: The specific steps for obtaining the performance coefficients of the training model are as follows: The sensor data units and order of the feature list are predefined. Preprocessed sensor data are concatenated according to the feature list. For each sensor data point, the maximum and minimum values ​​are iterated, and the data are mapped to values ​​from 0 to 1 using the maximum and minimum values. The normalized feature vectors are used to construct the input vector, which is then divided into training, validation, and test sets. A multiple linear regression model is selected for training. This model characterizes the linear relationship between the dependent variable and multiple independent variables and fits the influence of different independent variables on the dependent variable. Hyperparameters in the model are enumerated and selected. The selected candidate values ​​are then used to form a Cartesian product to form a network for evaluation. The root mean square error (RMSE) between the predicted values ​​of the multiple linear regression model and the actual data in the validation set is calculated using the following formula: error squared Let N represent the root mean square error, N represent the number of samples selected, and j represent the j-th sample. Let y represent the predicted value of the j-th sample. i This represents the true value of the j-th sample.

6. The control method for a dual-source heat pump system according to claim 5, characterized in that: The specific steps for obtaining performance coefficients from the training model also include: Candidate values ​​for hyperparameters are selected based on the root mean square error. The model is trained using the selected hyperparameters and the training set. The absolute error between the predicted and actual values ​​is defined as the loss function. The loss value for each sample in the training set is calculated, and the average value is taken as the training error. The collected data in the training set is constructed into a matrix. The regression coefficient vector is obtained using the least squares method. A set of undetermined regression coefficient vectors is set. The regression coefficient vectors and the constructed matrix of the collected data are used for matrix calculation. When the training error is set to 0, the partial derivatives of the regression coefficient vectors are calculated. The gradient of the partial derivatives of the regression coefficient vectors is set to 0 and solved.

7. The control method for a dual-source heat pump system according to claim 1, characterized in that: The specific steps for generating decisions using the design scoring mechanism are as follows: The predicted COP is obtained using multiple linear regression models for both ground-source and air-source heat sources. The energy consumption and mechanical wear caused by a single switch are converted into consumption costs. A comprehensive score function is calculated for the two heat sources. score a =α1COP a -α2cost switch -α3ΔT a ,a∈{air,ground}, Among them, score a This represents the overall score of the a-th heat source, where α1 represents the performance coefficient weight, α2 represents the cost weight, α3 represents the temperature difference weight, and COP is... a Let cost represent the COP of the a-th heat source. switch ΔT represents the cost of switching heat sources. a This represents the temperature difference of the heat exchanger for the a-th heat source, where air represents the air source and ground represents the ground source.

8. The control method for a dual-source heat pump system according to claim 7, characterized in that: The specific steps for generating decisions using the design scoring mechanism also include: A switching interval is set, and the current scores of different heat sources are calculated in real time. First, the switching interval is judged. If the time since the last switching is less than the set interval, no switching operation is performed. If the set interval is exceeded, a decision to switch heat sources is generated based on the scores. air >score ground When switching to the air source, when the score air <score ground When switching to ground source, when score air =score ground At that time, leave the current heat source unchanged.

9. The control method for a dual-source heat pump system according to claim 1, characterized in that: The specific steps for the smooth switching of the decision-making process are as follows: After the edge gateway obtains the decision, it sends the decision to the field execution unit to carry out preheating preparation. First, it calculates the temperature difference between the two sources and heats up or cools down the heat source to be switched. When the temperature difference between the two sources is close, it adjusts the valve ratio to perform dynamic load compensation, collects the temperature difference value, and adjusts the compensation amount according to the difference value, increasing or decreasing the pump speed and fan speed. It also collects subsequent difference values ​​at regular intervals to continue adjustment. After the valve adjustment and dynamic load compensation are completed, the switching status is determined, the valve of the previous heat source is closed, and the pipeline of the previous heat source is disconnected.

10. The control method for a dual-source heat pump system according to claim 1, characterized in that: The specific steps for periodically collecting feedback and iterative optimization are as follows: Periodically export decision logs from BMS, including COP predicted and actual values, switch heat source types for decision-making, and calculate COP error using the following formula: in This represents the performance coefficient error of the k-th data point. COP represents the prediction performance coefficient for the k-th data point. k This represents the actual performance coefficient of the k-th data item; The mean and variance of COP error are calculated, classified by ground source and air source, and the deviation of different heat sources is analyzed. Warning thresholds are set for the mean and variance, and automatic alarms are issued for indicators that exceed the thresholds. The assessment results and indicators that exceed the thresholds are sent to the operation and maintenance department via SMS. The weights in the comprehensive scoring function are fine-tuned based on the warning information, error data is collected, and the data is input into the multiple linear regression model for retraining, thereby updating the parameters.

Images