Multi-control object modeling method for air conditioning system

Abstract

The invention discloses an air conditioning system-oriented multi-control object modeling method, which comprises the following steps of: acquiring multi-source operation data acquired by an air conditioning system sensor, performing principal component analysis and time shift correlation analysis on the multi-source operation data, and screening out a core input variable corresponding to a target control object; the target control object comprises virtual indoor temperature, air source heat pump unit energy consumption, water pump energy consumption and fan energy consumption; according to the time-shifting correlation analysis, the lag step length of each variable is determined through the maximum information coefficient, and key time sequence features are screened; and on the basis of the screened core input variables, data driving models of all the control objects are constructed, and the combined type multi-control-object simulation framework is used for predicting the energy consumption and the environment state of the air conditioning system. According to the invention, the dependence of a test method on an actual system and hardware is overcome, and the test cost and risk are reduced. The source code can be tested before system and hardware configuration, and the problem of frequent equipment control caused by testing in an actual system is avoided.

Description

Technical Field

[0001] The present invention relates to a multi - control - object modeling method for an air - conditioning system. Background Art

[0002] The realization of the whole - process automation, informatization, digitalization and advanced control technology of the air - conditioning system is an important direction for the intelligent development of modern buildings. In this process, software integrating relevant intelligent algorithms plays a crucial role. However, most traditional software testing methods rely on actual systems and hardware for testing, such as hardware - in - the - loop simulators, etc. This method has many limitations in practical applications.

[0003] Especially for the development stage of early software and related control algorithms, it is almost unrealistic to use the actual system for testing. On the one hand, the air - conditioning system of large public buildings is complex and huge. During testing, frequent control of equipment is required, which will cause great interference to the stable operation of the system. On the other hand, the actual test time is generally short and discontinuous, making it difficult to comprehensively verify the functions and performance of the software.

[0004] The invention patent with the publication number CN110673585B proposes an innovative testing method for the train air - conditioning system. This method first constructs a train control system TCMS simulation device and an air - conditioning system simulation device, and designs corresponding air - conditioning condition strategies. Subsequently, according to the input test configuration parameters, the system calls and integrates the configuration items in the preset air - conditioning condition strategies to automatically generate test cases. These test cases will be transmitted to the train control system TCMS simulation device, which tests the air - conditioning system simulation device according to the test cases. Through the above - mentioned simulation devices, the testing of the train air - conditioning system can be realized, providing ideas for the testing of the station air - conditioning system. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art, overcome the dependence of traditional testing methods on actual systems and hardware, reduce testing costs and risks, and provide a multi - control - object modeling method for an air - conditioning system.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A multi - control - object modeling method for an air - conditioning system includes: obtaining multi - source operation data collected by sensors of the air - conditioning system, where the multi - source operation data includes outdoor temperature (To), outdoor relative humidity (Ho), indoor temperature (Tn), supply water temperature (Tsup), return water temperature (Tret), supply water flow (msup), pump frequency (npump), and the number of indoor persons (Np); performing principal component analysis (PCA) and time - shift correlation analysis on the multi - source operation data to screen out core input variables corresponding to the target control object; wherein,

[0008] The target control objects include virtual indoor temperature, energy consumption of air source heat pump units, energy consumption of water pumps, and energy consumption of fans;

[0009] The time-lagged correlation analysis determines the lag steps of each variable through the maximum information coefficient (MIC) to screen key time series features; based on the screened core input variables, data-driven models for each control object are constructed respectively, and the models are trained and verified to form a combined multi-control object simulation framework; wherein, the combined multi-control object simulation framework is used to predict the energy consumption and environmental status of the air conditioning system and output them to the model predictive control (MPC) optimizer.

[0010] As a preferred method, the steps of constructing the virtual indoor temperature model include: determining input variables through MIC analysis, including the lag 0-3 step features of outdoor temperature (To), the lag 0-2 step features of outdoor humidity (Ho), and the current number of indoor people (Np); performing PCA dimensionality reduction on the outdoor temperature and humidity, and taking the principal components with a cumulative variance contribution rate ≥ 98% as the input variables after dimensionality reduction, denoted as Toa1, Toa2, Hoa1, and Hoa2 respectively; combining the variables after dimensionality reduction with historical supply water temperature (Tsup0 - Tsup1), historical supply water flow (msup0 - msup1), and current indoor temperature (Tn) to construct a training set with 10 input variables; using the Att-BiGRU algorithm to model the training set and optimizing the network parameters through backpropagation to obtain a virtual indoor temperature simulation model with a prediction error of MAE ≤ 0.07 and RMSE ≤ 0.12.

[0011] As a preferred method, the steps of constructing the energy consumption model of the air source heat pump unit include: screening input variables, including the current supply water temperature (Tsup), return water temperature (Tret), flow rate (msup), outdoor temperature (To), outdoor humidity (Ho), indoor people (Np), indoor temperature (Tn), and historical cooling capacity (Qhis); implanting the input variables into a stacked wide learning system to establish an energy consumption prediction sub-model, and its output is the coefficient of performance (COP) of the heat pump, and the calculation relationship satisfies: COP = f(T low , T high ), where T low is the outlet water temperature and T high is the outdoor temperature; integrating the cooling capacity sub-model and the COP sub-model, and outputting the host energy consumption value through the formula W ASHP = Q ASHP / COP * ΔT, where Q is the system cooling capacity per unit time, kW; ΔT is the system operation time, h.

[0012] As a preferred method, when constructing the water pump energy consumption model, the following operations are performed:

[0013] The flow rate of the water pump is proportional to the frequency, and the electric power is proportional to the cube of the frequency. The relationships between the water pump flow rate m and the electric power Q and the water pump frequency n are shown in Equations (1-1) and (1-2):

[0014]

[0015] Where: m1 and m0 are the flow rates of the water pump under actual and rated operating conditions respectively, m 3 / h; n1 and n0 are the frequencies of the water pump under actual and rated operating conditions respectively, Hz; Q1 and Q0 are the electric powers of the water pump under actual and rated operating conditions respectively, kW;

[0016] For a water system where the water flow rate of the terminal load can be actively adjusted, due to the change of the pipe network characteristic curve, the similar operating condition is no longer satisfied. And as the water pump flow rate decreases, while the input power decreases, the water pump efficiency will decrease sharply. The change of the final energy consumption is affected by many factors such as the water pump model, valve characteristics, control mode, etc., and it is difficult to accurately describe. According to the literature, the relationships between the flow rate and the electric power of the water pump under actual operating conditions and the frequency can be expressed by Equations (1-3) and (1-4).

[0017]

[0018] Where: a, b, c, d, e, f are coefficients fitted by measured data.

[0019] As a preferred method, a combined air handling unit model is constructed, including establishing a return water temperature prediction model:

[0020] For a constant air volume primary return air air conditioning system, the initial state dry bulb temperature of the air for air treatment can be obtained by mixing indoor air and outdoor fresh air, which is t1, the enthalpy value is h1, the wet bulb temperature is t s1 , the air volume to be treated is G, the chilled water flow rate is m, and the initial temperature of the chilled water is T sup , then the final state t2 and the return water temperature T ret that the surface cooler can reach for treating air are as follows:

[0021] A1. Calculate the face velocity V y and the water velocity w of the surface cooler, and the calculation formulas are shown in Equations (2-1) and (2-2):

[0022]

[0023] Where: F y is the face area of the surface cooler, m 2 ; ρ is the density of water, kg / m 3 ; f w is the water passing cross-sectional area, m2 。

[0024] A2. Determine the general heat exchange efficiency E' that the surface cooler can provide. This value can be obtained from the equipment sample.

[0025] A3. Assume that the dry-bulb temperature of the air at the final state is t2.

[0026] It can be determined that the dry-bulb temperature of the air at the final state is t2 = t w1 +(4 - 6)°C, where t w1 is the measured value of the first-stage water supply temperature in the air-conditioning system. The wet-bulb temperature t s2 at the final state is calculated as shown in Equation (2-3):

[0027] t s2 = t2 - (t1 - t s1 )(1 - E') (2-3)

[0028] The enthalpy value h2 at the final state is calculated as shown in Equation (2-4):

[0029]

[0030] A4. Calculate the dehumidification coefficient ξ, and the calculation formula is as shown in (2-5):

[0031]

[0032] A5. Calculate the heat transfer coefficient K s 。

[0033] The empirical formula for the heat transfer coefficient of the heat exchanger in the dehumidifying and cooling process is as shown in Equation (2-6):

[0034]

[0035] Where: A, P, B, m, n are coefficients and exponents obtained from experiments.

[0036] A6. Calculate the total heat exchange efficiency E' that the surface cooler can achieve g , and the calculation formula is as shown in Equations (2-7) - (2-9):

[0037]

[0038] A7. Calculate the required total heat exchange efficiency E g and compare it with E' g , and the calculation formula of E g is as shown in Equation (2-10):

[0039]

[0040] When |E g - E'g When |≤δ (generally, 0.01 can be taken), it is proved that the assumed t2 is appropriate. If |E g - E' g |>δ, then t2 should be reset and the calculation should be carried out again.

[0041] A8. Calculate the return water temperature T of the chilled water ret , and the calculation formula is shown in Equation (2 - 11).

[0042]

[0043] As an optimal method, a combined air handling unit model is constructed, including: establishing a return water temperature prediction model: through MIC analysis, the input variables are determined to be To lag 1 step, Ho lag 1 step, Tn lag 1 step, Tsup lag 1 step, and msup lag 1 step, and an Att - BiGRU algorithm is used to train a return water temperature prediction sub - model with MAE ≤ 0.11; establishing a fixed - frequency fan energy consumption model: the energy consumption of the combined air handling unit is the fan energy consumption. According to the similarity law, the power of the fan is proportional to the cube of its operating frequency. The fan in the combined air handling unit is a fixed - frequency fan, and its operating air volume and power remain unchanged; the fan energy consumption W fan The calculation formula is shown in Equation (3 - 1):

[0044] W fan = Q fan *ΔT (3 - 1)

[0045] In the formula: Q fan is the actual operating power of the fan, in kW.

[0046] As an optimal method, the model training and verification steps include: dividing the sample data into a training set and a verification set at a ratio of 7:3, using the first 10 time steps as input to predict the output of the next time step; using the early stopping method to monitor the verification set error, and terminating the training when the RMSE decreases by less than 1% for 10 consecutive iterations; calculating the mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE) of the verification set, and selecting Att - BiGRU as the optimal basic algorithm, and optimizing its model weights through gradient descent.

[0047] As an optimal method, it also includes a model integration method based on a simulation platform: building a building data model in the Matlab environment, establishing a Modbus / OPC communication link with the MPC controller through the kepserverex software; integrating the virtual indoor temperature model, air source heat pump model, water pump model, and air handling unit model in parallel, and ensuring the consistency of the simulation step size through a time synchronization module; generating model output signals including: real - time temperature prediction value Tn + 1, host energy consumption W ASHP, the energy consumption W of the water pump pump , the energy consumption W of the fan fan , and map it to the register address of KepserverEX for the controller to read.

[0048] As a preferred method, the construction of the communication link specifically includes: creating a virtual device in KepserverEX, binding the input and output variables of the Matlab model to the Modbus TCP protocol register; setting the simulation step size to 5 minutes, and completing the closed-loop process of data writing → model calculation → register update → controller reading within each step; configuring an OPC UA client on the controller side to regularly read the predicted temperature and energy consumption data in the register and feedback it to the MPC optimization algorithm.

[0049] As a preferred method, it also includes a model online update method: deploying an operation data monitoring module to collect the actual temperature value T_real and the model prediction value T_pred in real time; when the daily average prediction error exceeds the threshold e (e = 0.5 °C), trigger the model retraining process; adopt an incremental learning method, add the latest 7-day data to the training set, update the weights of the Att-BiGRU network, and maintain the model prediction accuracy.

[0050] The present invention has at least the following beneficial effects: The present invention overcomes the dependence of traditional testing methods on actual systems and hardware, and reduces the testing cost and risk. By constructing a data-driven combined multi-control object simulation framework, it is possible to test the source code before the system and hardware configuration, avoiding problems such as frequent device control and interference with the stable operation of the system caused by testing in the actual system. The present invention improves the testing efficiency and comprehensiveness. Through principal component analysis and time-shift correlation analysis, the core input variables corresponding to the target control object are screened out, and an efficient data-driven model is constructed. This model can comprehensively simulate the operating state of the air-conditioning system, providing strong support for the automatic generation and testing of test cases, thereby improving the testing efficiency and comprehensiveness. Brief Description of the Drawings

[0051] In order to reveal the technical details of the embodiments of the present invention, the drawings involved in the embodiments will be briefly introduced next. It should be emphasized that these drawings only present several embodiments of the present invention and should not be regarded as a definition of the scope of the invention. For those of ordinary skill in the art, other related drawings can still be derived based on these drawings without creative labor.

[0052] Figure 1 Schematic diagram of the communication method between sensors, devices and the software platform in the embodiment;

[0053] Figure 2 Simulation environment of the station air-conditioning control system in the embodiment;

[0054] Figure 3 The MIC values of each influencing factor and the indoor temperature in the embodiments;

[0055] Figure 4 The MIC values of each influencing factor and the return water temperature in the embodiments. Specific implementation manners

[0056] The technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the following description.

[0057] In the following content, the embodiments of the present disclosure will be described in detail with the aid of the accompanying drawings. However, it should be clear that the present disclosure is not limited to the specific forms shown here. On the contrary, it should be understood to cover various variations, equivalent forms, and / or alternative solutions of the embodiments of the present disclosure. During the description of the drawings, the same reference numerals will be used to label similar components.

[0058] In each embodiment of the present disclosure, expressions such as "first", "second", "the first", or "the second" are used to modify different components, rather than indicating order and / or importance, nor imposing limitations on the corresponding components. For example, the first user device and the second user device respectively represent different user devices, although they both belong to the category of user devices. Similarly, the first component can be named the second component, and the second component can also be named the first component, which does not change their essential attributes within the scope of the present disclosure.

[0059] In this disclosure, terms are used to illustrate specific embodiments and do not constitute a limitation to the present disclosure. In this context, the use of the singular form also covers the plural form unless otherwise clearly stated in the text. During the description, it should be understood that terms such as "including" or "having" are intended to indicate the existence of features, quantities, steps, operations, structural components, parts, or combinations thereof, without precluding the possibility or addition of one or more other features, quantities, steps, operations, structural components, parts, or combinations thereof.

[0060] It should be clear that although detailed specific details are provided in the following description to help comprehensively understand the example embodiments. However, professionals in the field should know that even without these specific details, the example embodiments can still be implemented. For example, the system can be presented in the form of a block diagram to avoid excessive details interfering with the clarity of the example. In other cases, to maintain the clarity of the example, unnecessary details of those well-known processes, structures, and technologies may be omitted.

[0061] Multi-control object modeling method for air conditioning system, including: obtaining multi-source operation data collected by sensors of the air conditioning system, where the multi-source operation data includes outdoor temperature (To), outdoor relative humidity (Ho), indoor temperature (Tn), supply water temperature (Tsup), return water temperature (Tret), supply water flow rate (msup), pump frequency (npump), and number of indoor occupants (Np); performing principal component analysis (PCA) and time-shift correlation analysis on the multi-source operation data to screen out core input variables corresponding to the target control object; where,

[0062] The target control objects include virtual indoor temperature, energy consumption of air source heat pump units, pump energy consumption, and fan energy consumption;

[0063] The time-shift correlation analysis determines the lag step of each variable through the maximum information coefficient (MIC) and screens out key time series features; based on the screened core input variables, dynamic data-driven models of each control object are constructed respectively, and the models are trained and verified through a time series algorithm enhanced by the attention mechanism to form a combined multi-control object simulation framework; where, the combined multi-control object simulation framework is used to predict the energy consumption and environmental state of the air conditioning system and output them to the model predictive control (MPC) optimizer.

[0064] The multi-control object modeling method for air conditioning system provides comprehensive and accurate data support for subsequent modeling by obtaining multi-source operation data collected by sensors of the air conditioning system, including outdoor temperature, outdoor relative humidity, indoor temperature, supply water temperature, return water temperature, supply water flow rate, pump frequency, and number of indoor occupants, etc. By performing principal component analysis (PCA) and time-shift correlation analysis on the multi-source operation data to screen out core input variables corresponding to the target control object, the complexity of the model is effectively reduced and the operation efficiency of the model is improved. At the same time, the time-shift correlation analysis determines the lag step of each variable through the maximum information coefficient (MIC) and screens out key time series features, further enhancing the prediction ability of the model. Based on the screened core input variables, dynamic data-driven models of each control object are constructed respectively, and the models are trained and verified through a time series algorithm enhanced by the attention mechanism to form a combined multi-control object simulation framework. This framework can accurately predict the energy consumption and environmental state of the air conditioning system, providing strong support for the subsequent model predictive control (MPC) optimizer. This method not only improves the operation efficiency of the air conditioning system, reduces energy consumption, but also optimizes the indoor environmental state and improves the comfort of users. At the same time, this method has good generality and scalability, can be applied to different types of air conditioning systems and control objects, and provides new ideas and methods for the intelligent management and optimization of air conditioning systems.

[0065] The present invention overcomes the dependence of traditional testing methods on actual systems and hardware, reducing testing costs and risks. By constructing a data-driven combined multi-control object simulation framework, it is possible to test the source code before the system and hardware are configured, avoiding problems such as frequent device control and interference with the stable operation of the system during testing in the actual system. The present invention improves testing efficiency and comprehensiveness. Through principal component analysis and time-shifted correlation analysis, the core input variables corresponding to the target control object are screened out, and an efficient data-driven model is constructed. This model can comprehensively simulate the operating state of the air-conditioning system, providing strong support for the automatic generation and testing of test cases, thereby improving testing efficiency and comprehensiveness.

[0066] This embodiment can effectively verify the predictive control algorithm of the air-conditioning system. The proposed simulation platform can simulate various working conditions and conditions, verify and optimize the predictive control algorithm of the air-conditioning system, improve the stability and accuracy of the algorithm, and provide strong support for the research and development of air-conditioning system software. In summary, the proposed multi-control object modeling method for air-conditioning systems effectively overcomes the limitations of traditional testing methods, improves testing efficiency and comprehensiveness, realizes the effective verification of the predictive control algorithm of air-conditioning systems, and provides strong support for the research and development of air-conditioning system software.

[0067] In one embodiment, the steps of constructing a virtual indoor temperature model include: determining input variables through MIC analysis, including the lag 0-3 step characteristics of the outdoor temperature (To), the lag 0-2 step characteristics of the outdoor humidity (Ho), and the current number of indoor people (Np); performing PCA dimensionality reduction processing on the outdoor temperature and humidity, and taking the principal components with a cumulative variance contribution rate ≥ 98% as the input variables after dimensionality reduction, denoted as Toa1, Toa2 and Hoa1, Hoa2 respectively; combining the variables after dimensionality reduction with the historical water supply temperature (Tsup0 - Tsup1), historical water supply flow rate (msup0 - msup1), and current indoor temperature (Tn) to construct a training set with 10 input variables; using the Att-BiGRU algorithm to model the training set, and optimizing the network parameters through backpropagation to obtain a virtual indoor temperature simulation model with a prediction error MAE ≤ 0.07 and RMSE ≤ 0.12.

[0068] In this embodiment, the input variables are determined through the maximum information coefficient (MIC) analysis. These variables include the lag 0-3 step features of the outdoor temperature (To), the lag 0-2 step features of the outdoor humidity (Ho), and the current number of indoor people (Np). These variables will serve as the basic data for constructing the model. Principal component analysis (PCA) dimensionality reduction processing is performed on the outdoor temperature and humidity. During the PCA process, the principal components with a cumulative variance contribution rate greater than or equal to 98% are selected as the input variables after dimensionality reduction. These principal components will be denoted as Toa1, Toa2 (representing the two principal components after dimensionality reduction of the outdoor temperature) and Hoa1, Hoa2 (representing the two principal components after dimensionality reduction of the outdoor humidity). The variables after dimensionality reduction are combined with the historical water supply temperature (Tsup0-Tsup1, that is, the water supply temperatures at the current and previous moments), the historical water supply flow rate (msup0-msup1, that is, the water supply flow rates at the current and previous moments), and the current indoor temperature (Tn). In this way, a training set with 10 input variables is constructed. The attention mechanism bidirectional gated recurrent unit (Att-BiGRU) algorithm is used to model the training set. The network parameters are optimized through the backpropagation algorithm so that the model can accurately predict the indoor temperature. After training and optimization, the obtained virtual indoor temperature simulation model should satisfy that the mean absolute error (MAE) of the prediction error is less than or equal to 0.07, and the root mean square error (RMSE) is less than or equal to 0.12. It can comprehensively consider multiple factors such as outdoor temperature, humidity, the number of indoor people, and water supply temperature and flow rate, and achieve accurate prediction of indoor temperature. This not only helps to improve the comfort of the indoor environment but also provides strong support for the energy management and energy conservation of buildings.

[0069] In one embodiment, the steps of constructing an air source heat pump unit energy consumption model include: screening input variables through a fuzzy clustering algorithm, including the current water supply temperature (Tsup), return water temperature (Tret), flow rate (msup), outdoor temperature (To), outdoor humidity (Ho), indoor occupants (Np), indoor temperature (Tn), and historical cooling capacity (Qhis); implanting the input variables into a stacked wide learning system to establish an energy consumption prediction sub-model, and its output is the coefficient of performance (COP) of the heat pump, and the calculation relationship satisfies: COP = f(T low ,T high ), that is where T low is the outlet water temperature, and T high is the outdoor temperature; the output main engine energy consumption value is output through the formula W ASHP = Q ASHP / COP*ΔT, where Q ASHP is the system cooling capacity per unit time, in kW; ΔT is the system operation time, in h.

[0070] The energy consumption of the host is related to the system cooling capacity and COP. For air source heat pump units, under the refrigeration condition, the system energy consumption is related to the supply water temperature T sup , the return water temperature T ret , the flow rate m sup , the outdoor ambient temperature T o , the relative humidity H o , the number of people N p , the indoor temperature T N and the historical cooling capacity Q his . Therefore, the host energy consumption model can be expressed as shown in Equation (1):

[0071] W ASHP = f ASHP (T sup , T ret , m sup , T o , H o , N p , T N , Q his ) * ΔT (1)

[0072] In the formula: f ASHP is a data-driven model related to the energy consumption calculation of air source heat pump units. In this embodiment, the input variables are screened by the fuzzy clustering algorithm, and the host energy consumption model is established by using the Stacked Broad Learning System.

[0073] Fuzzy clustering variable screening process: Data preprocessing: Standardize the input variables (such as the supply water temperature T sup , the return water temperature T ret , the flow rate m sup , the outdoor ambient temperature T o , the relative humidity H o , the number of people N p , the indoor temperature T N and the historical cooling capacity Q his ). Algorithm selection: Adopt fuzzy C-means clustering (FCM), define the membership function as Gaussian type, and optimize the number of clusters through the Silhouette Index. Variable screening: Screen the key variables according to the membership degree (>0.7) of the variables in the cluster center and the mutual information with the target variable Q. Verification: Compare the model performance (such as RMSE, MAE) before and after screening through cross-validation. This embodiment avoids the interference of redundant variables, reduces the model complexity, and improves the generalization ability.

[0074] Stacked width learning system modeling process: Model structure: Input layer: Receives the screened variables. Feature node layer: Generates initial feature nodes (e.g., 100) through random mapping. Enhanced node layer: Nonlinear transformation (e.g., Sigmoid) expands features and enhances the expressive ability. Stacked design: Multiple layers of feature-enhanced nodes are stacked to extract high-order features layer by layer (e.g., 3-layer stack). Output layer: Linear regression predicts energy consumption. Training strategy: Incremental learning: Dynamically expands nodes to adapt to changes in data distribution. Regularization: Adopts L2 regularization to prevent overfitting, and hyperparameters are optimized through grid search. The stacked structure enhances the nonlinear fitting ability and is suitable for energy consumption prediction with multi-variable coupling.

[0075] In one embodiment, when constructing a water pump energy consumption model, the following operations are performed: The flow rate of the water pump is proportional to the frequency, and the electric power is proportional to the cube of the frequency; the relationships between the water pump flow rate m and the electric power Q and the water pump frequency n are shown in Equations (1-1) and (1-2):

[0076]

[0077] where: m1 and m0 are the flow rates of the water pump under actual and rated working conditions respectively, m 3 / h; n1 and n0 are the frequencies of the water pump under actual and rated working conditions respectively, Hz; Q1 and Q0 are the electric powers of the water pump under actual and rated working conditions respectively, kW;

[0078] For a water system where the end-load water flow rate can be actively adjusted, since the characteristic curve of the pipe network will change, the similar working condition conditions are no longer satisfied. And as the water pump flow rate decreases, while the input power decreases, the water pump efficiency will decrease sharply. The change of the final energy consumption is affected by many factors such as the water pump model, valve characteristics, and control method, and it is difficult to accurately describe. According to the literature, the relationships between the flow rate and electric power of the water pump under actual working conditions and the frequency can be expressed by Equations (1-3) and (1-4).

[0079]

[0080] where: a, b, c, d, e, f are coefficients fitted through measured data;

[0081] According to the collected data, the relationships between the flow rate and electric power of the water pump under actual working conditions and the frequency are shown in Equations (1-5) and (1-6).

[0082] m pump = 6.5001n pump - 40.5961 (1-5)

[0083] Q pump = -0.0083n pump3 +1.0829n pump 2 -45.6339n pump +641.4388 (1 - 6)

[0084] Where: m pump is the flow rate of the water pump in the actual working condition of this engineering case, m 3 / h; n pump is the frequency of the water pump in the actual working condition, Hz; Q pump is the electric power of the water pump in the actual working condition, kW.

[0085] The energy consumption W of the water pump pump is calculated as shown in Equation (1 - 7):

[0086] W pump = Q pump *ΔT(1 - 7).

[0087] In another embodiment, a combined air handling unit model is constructed, including establishing a return water temperature prediction model:

[0088] The combined air handling unit mainly consists of equipment such as surface coolers, humidifiers, heaters, and fans. During the refrigeration season, the surface cooler is mainly used to cool and dehumidify the air. For a constant air volume air conditioning system with primary return air, the mixed indoor air and outdoor fresh air can obtain the initial state of the air treatment with a dry bulb temperature of t1, an enthalpy value of h1, and a wet bulb temperature of t s1 , the air volume to be processed is G, the chilled water flow rate is m, and the initial temperature of the chilled water is T sup , then the final state t2 that the surface cooler can reach for processing the air and the return water temperature T ret are as follows:

[0089] A1. Calculate the face velocity V y of the surface cooler and the water flow velocity w, and the calculation formulas are as shown in Equations (2 - 1) and (2 - 2):

[0090]

[0091] Where: F y is the frontal area of the surface cooler, m 2 ; ρ is the density of water, kg / m 3 ; f w is the cross - sectional area for water flow, m 2 .

[0092] A2. Obtain the general heat transfer efficiency E' that the surface cooler can provide, and this value can be obtained from the equipment sample.

[0093] A3. Assume that the dry bulb temperature of the final air state is t2.

[0094] The dry-bulb temperature of the final air state can be determined according to t2 = t w1 +(4 - 6)°C, where t w1 is the measured value of the first-stage water supply temperature in the air-conditioning system. The wet-bulb temperature t of the final state s2 is calculated as shown in Equation (2-3):

[0095] t s2 = t2 - (t1 - t s1 )(1 - E') (2-3)

[0096] The enthalpy value h2 of the final state is calculated as shown in Equation (2-4):

[0097]

[0098] A4. Calculate the moisture extraction coefficient ξ, and the calculation formula is as shown in (2-5):

[0099]

[0100] A5. Calculate the heat transfer coefficient K s .

[0101] The empirical formula for the heat transfer coefficient of the dehumidifying and cooling process of the heat exchanger is as shown in Equation (2-6):

[0102]

[0103] In the formula: A, P, B, m, n are coefficients and exponents obtained from experiments.

[0104] A6. Calculate the total heat exchange efficiency E' that the surface cooler can achieve g , and the calculation formula is as shown in Equations (2-7) - (2-9):

[0105]

[0106] A7. Calculate the required total heat exchange efficiency E g and compare it with E' g . The calculation formula of E g is as shown in Equation (2-10):

[0107]

[0108] When |E g - E' g | ≤ δ (generally, 0.01 can be taken), it is proved that the assumed t2 is appropriate. If |E g - E' g | > δ, then t2 should be reset and recalculated.

[0109] A8. Calculate the return water temperature T of chilled water ret , and the calculation formula is as shown in Equation (2-11).

[0110]

[0111] Through the above steps, we can calculate the dry bulb temperature t2 at the final state that the air handling unit can reach and the return water temperature Tret of chilled water. This method not only improves the accuracy of air handling but also optimizes the energy usage efficiency. Through detailed calculations, it can ensure that the air reaches the expected dry bulb temperature t2, wet bulb temperature ts2, and enthalpy value h2 after treatment, meeting the specific requirements of the air conditioning system. During the calculation process, multiple factors such as the heat exchange efficiency, heat transfer coefficient, and moisture extraction coefficient of the air handling unit are considered, so that the flow rate and temperature of chilled water can be more reasonably configured, reducing unnecessary energy consumption. By assuming and verifying the dry bulb temperature at the final state of the air, the stability of the system during actual operation can be ensured. When there is a deviation between the assumed t2 and the actual demand, adjustments can be made in a timely manner to avoid excessive or too small fluctuations in the system. This embodiment enables system maintenance personnel to more clearly understand the operating conditions of the system, discover and solve problems in a timely manner, reducing the system's maintenance cost and management difficulty.

[0112] In one embodiment, a combined air handling unit model is constructed, including: establishing a return water temperature prediction model in another way: determining the input variables as To lag 1 step, Ho lag 1 step, Tn lag 1 step, Tsup lag 1 step, and msup lag 1 step through MIC analysis, and training with the Att-BiGRU algorithm to obtain a return water temperature prediction sub-model with MAE ≤ 0.11; establishing a fixed-frequency fan energy consumption model: the energy consumption of the combined air handling unit is the fan energy consumption. According to the similarity law, the power of the fan is proportional to the cube of its operating frequency. The fan in the combined air handling unit is a fixed-frequency fan, and its operating air volume and power remain unchanged; the fan energy consumption W fan The calculation formula is as shown in Equation (3-1):

[0113] W fan =Q fan *ΔT (3-1)

[0114] Where: Q fan is the actual operating power of the fan, in kW.

[0115] The total energy consumption calculation formula of the air conditioning system is obtained as shown in Equation (3-2):

[0116] W total =(f ASHP +Q pump +Q fan )*ΔT (3-2)

[0117] In this embodiment, through MIC analysis, we determined that the input variables of the return water temperature prediction model are the outdoor temperature (To) lagged by 1 step, the indoor humidity (Ho) lagged by 1 step, the indoor temperature (Tn) lagged by 1 step, the supply water temperature (Tsup) lagged by 1 step, and the supply water flow rate (msup) lagged by 1 step. The Att-BiGRU algorithm was used for training, and a return water temperature prediction sub-model with a mean absolute error (MAE) less than or equal to 0.11 was obtained. This model can accurately predict the return water temperature, providing strong support for the energy efficiency management and optimization of the air conditioning system. When establishing the fixed-frequency fan energy consumption model, according to the similarity law and the actual situation of the fan in the packaged air handling unit, it was determined that the power of the fan is proportional to the cube of its operating frequency. Moreover, since the fan operates at a fixed frequency, its operating air volume and power remain unchanged. Through formula calculation, the actual operating power of the fan can be obtained, and then the total energy consumption of the air conditioning system can be obtained. This model provides an important basis for evaluating and optimizing the energy consumption of the air conditioning system. The constructed packaged air handling unit model can accurately predict the return water temperature and effectively evaluate the energy consumption of the air conditioning system, providing strong support for the efficient operation and energy-saving management of the air conditioning system.

[0118] In one embodiment, when constructing a virtual indoor temperature model or establishing a return water temperature prediction model, the model training and validation steps include: dividing the sample data into a training set and a validation set at a ratio of 7:3, using the first 10 time steps as input to predict the output of the next time step; using the early stopping method to monitor the validation set error, and terminating the training when the RMSE decrease is less than 1% for 10 consecutive iterations; calculating the mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE) of the validation set, and selecting Att-BiGRU as the optimal basic algorithm, whose model weights are optimized by gradient descent. By dividing the sample data into a training set and a validation set at a ratio of 7:3, the present invention ensures that the model can fully learn the internal laws of the data during the training process and effectively evaluate the generalization ability of the model through the validation set, avoiding the occurrence of overfitting. The method of using the first 10 time steps as input to predict the output of the next time step makes full use of the autocorrelation of time series data and improves the prediction accuracy of the model for temperature change trends. Using the early stopping method to monitor the validation set error and terminating the training when the RMSE decrease is less than 1% for 10 consecutive iterations, this strategy effectively prevents overtraining of the model during the training process, saves computing resources, and at the same time ensures the generalization performance of the model. By calculating the mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE) of the validation set, the prediction performance of the model can be comprehensively evaluated, providing a reliable basis for the selection and optimization of the model. Selecting Att-BiGRU as the optimal basic algorithm makes full use of its powerful modeling ability for time series data by bidirectional gated recurrent units and the ability of the attention mechanism to capture key information, thereby improving the prediction accuracy and stability of the model. The model weights are optimized by gradient descent, ensuring that the model can gradually approach the optimal solution during the training process and further improving the prediction performance of the model. The training and validation steps of the present invention for constructing a virtual indoor temperature model or a return water temperature prediction model significantly improve the prediction accuracy and generalization ability of the model through scientific and reasonable data set division, selection of effective prediction strategies, adoption of advanced algorithms and optimization methods, providing strong support for the research and application in the field of temperature prediction.

[0119] The present invention also includes a model integration method based on a simulation platform: building a building data model in the Matlab environment, establishing a Modbus / OPC communication link with the MPC controller through the kepserverex software; integrating the virtual indoor temperature model, air source heat pump model, water pump model, and air conditioner unit model in parallel, and ensuring the consistency of the simulation step size through the time synchronization module; generating model output signals including: real-time temperature prediction value Tn+1, main engine energy consumption W ASHP , water pump energy consumption W pump , fan energy consumption W fan, and map it to the register address of KepserverEX for the controller to read in real time.

[0120] The construction of the communication link specifically includes: creating a virtual device in KepserverEX, binding the input and output variables of the Matlab model to the Modbus TCP protocol register; setting the simulation step length to 5 minutes, and completing the closed-loop process of data writing → model calculation → register update → controller reading within each step; configuring an OPC UA client on the controller side to regularly read the predicted temperature and energy consumption data in the register and feedback it to the MPC optimization algorithm.

[0121] The present invention also includes a model online update method: deploying an operation data monitoring module to collect the actual temperature value T_real and the model prediction value T_pred in real time; when the daily average prediction error exceeds the threshold e (e = 0.5 °C), triggering the model retraining process; adopting an incremental learning method, adding the latest 7-day data to the training set, updating the weights of the Att-BiGRU network, and maintaining the model prediction accuracy. The application of this method in the central air-conditioning system of railway passenger stations includes: dividing the waiting hall into multiple thermal zones, performing virtual temperature modeling on each zone in parallel; dynamically correcting the input parameter Np based on the population density of each zone and synchronously adjusting the set value of the water pump frequency; calculating the optimal energy consumption strategy through the MPC controller and outputting the water supply temperature, flow rate, and equipment start-stop instructions to the on-site PLC actuator.

[0122] Corresponding to the method, the present invention also discloses a multi-control object modeling system for an air-conditioning system, including: a sensor data acquisition module for obtaining multi-source data of outdoor temperature, humidity, indoor temperature and humidity, and equipment operation frequency in real time; a feature processing module configured to perform principal component analysis (PCA) dimensionality reduction processing on the original data; screening key lag step features through the time-shifted MIC algorithm; a multi-model construction module including: a virtual temperature prediction unit based on Att-BiGRU; a host energy consumption prediction unit based on stacked width learning; a water pump energy consumption unit based on non-linear fitting; a fan energy consumption unit based on fixed parameters; a simulation integration platform for integrating sub-models and establishing an OPC communication interface through KepserverEX; an optimization control module for receiving the predicted temperature and energy consumption data output by the model and executing the MPC algorithm to generate optimization control instructions.

[0123] The feature processing module includes: an MIC analysis sub-unit for calculating the maximum information coefficient between variables and screening the lag step; a PCA dimensionality reduction sub-unit for performing eigen-decomposition on To and Ho and outputting the principal components after dimensionality reduction; a dynamic variable selection unit for associating Toa1, Toa2, Hoa1, and Hoa2 with the input of different device models based on thermodynamic rules.

[0124] The multi-model construction module further includes: a model verification unit configured to: divide the training set and the validation set; calculate the MAE, MAPE, and RMSE metrics; compare the performance of Att-LSTM, Att-GRU, and Att-BiLSTM, and select Att-BiGRU as the basic algorithm.

[0125] The simulation integration platform includes: a Matlab model container that runs virtual temperature prediction and equipment energy consumption calculation programs; a kepserverex communication middleware that configures a Modbus register mapping table; and a time synchronizer that triggers the model to recalculate every 5 minutes and updates the register data.

[0126] The present invention further includes an online update module, which includes: a data quality detection unit that identifies abnormal sensor data and triggers a cleaning process; an incremental learning unit that loads the latest data for retraining when the model prediction error exceeds the limit; and a version management unit that saves historical model parameters and supports rolling back to a stable version.

[0127] The present invention discloses a multi-control object modeling method for an air-conditioning system, including the following steps: collecting building environment data and air-conditioning equipment operation data, where the building environment data includes indoor temperature, outdoor temperature, humidity, personnel density, and solar radiation intensity, and the air-conditioning equipment operation data includes the host water supply temperature, return water temperature, flow rate, and equipment frequency; performing time-shift correlation analysis on the collected building environment data to determine the lag impact step of each environmental parameter on the target variable, and performing dimensionality reduction processing on the high-dimensional environmental parameters through principal component analysis to generate a set of characteristic input variables; constructing a bidirectional gated recurrent unit network model based on the attention mechanism, inputting the set of characteristic input variables and the air-conditioning equipment operation data into the model, training to obtain a virtual indoor temperature prediction model, and outputting the temperature prediction value and confidence interval for a preset future time period; establishing a compressor dynamic response model based on the operating characteristics of the air-conditioning host, and constructing a host energy consumption prediction function in combination with the equipment frequency and water supply flow data, where the function includes the product relationship of refrigerating capacity, performance coefficient, and time variable;

[0128] Establishing a cubic frequency-power correlation model for a variable-frequency water pump, and fitting a flow-frequency correction coefficient under the actual working condition based on the pipeline characteristic curve to generate a dynamic water pump energy consumption calculation equation;

[0129] Constructing a hybrid modeling framework for a combined air handling unit, including a heat transfer equation of a surface cooler based on physical mechanisms and a return water temperature prediction model based on data-driven methods, where the physical equation includes parameters such as face velocity, water flow velocity, and moisture pick-up coefficient;

[0130] Build a software-in-the-loop simulation platform, connect the building data model with the controller algorithm through a communication protocol, and integrate the virtual indoor temperature model, host model, water pump model and air handling unit model in the Matlab environment to form a co-simulation system; parallelize the discrete models into a unified framework through the Matlab integration platform, use the time synchronization module to ensure that the prediction step sizes of each model are consistent (such as a cycle of 5 minutes), and transmit the virtual temperature and energy consumption prediction results to the controller in real time through Modbus / OPC communication. For example, the heat pump energy consumption model predicts the host load based on the current supply water temperature and outdoor temperature, while the water pump model dynamically adjusts the flow rate and power based on the frequency change, and the output of the return water temperature model is fed back to the heat pump model to form a closed loop. The controller synthesizes all model outputs (such as temperature deviation from the predicted value, early warning of energy consumption surge), and dynamically adjusts the operating parameters of the air conditioning system (such as reducing the frequency to save energy), and finally realizes the joint optimization of temperature control and energy consumption.

[0131] Inject the actually collected disturbance signals into the simulation environment, conduct collaborative verification on the multi-control object models, calculate the root mean square error index of the predicted temperature and the actual temperature, and trigger the model parameter adaptive adjustment mechanism when the error exceeds the threshold;

[0132] Construct a predictive control optimization objective function based on the verified multi-object models, and the function includes temperature comfort weight, equipment energy consumption weight and control action smoothness constraint conditions; transmit the optimized control instructions to the physical device through the KepserverEX middleware to realize the closed-loop adjustment of the operating parameters of the air conditioning system.

[0133] The time-shift correlation analysis specifically includes: calculating the maximum information coefficient of each environmental parameter and the target temperature variable, and determining the lag time window with significant correlation; performing sliding window sampling on the outdoor temperature and humidity parameters to construct a multi-dimensional time series matrix; calculating the cumulative variance explanation rate of the eigenvalues of each dimension through principal component analysis, and selecting the principal components with an explanation rate exceeding 95% as the features after dimensionality reduction.

[0134] The construction of the bidirectional gated recurrent unit network model of the attention mechanism includes: introducing a time attention module after the bidirectional GRU layer to calculate the attention weights of the hidden states of each time step; designing a spatial attention subnet to dynamically allocate weights to the multi-variable input features; using a residual connection method to fuse the original features and the attention-weighted features to prevent gradient disappearance.

[0135] The construction process of the host energy consumption prediction function includes: collecting the COP measured data of the host at different load rates, and establishing an operating condition division model based on fuzzy C-means clustering; constructing a stacked width learning system prediction network in various operating condition subspaces, and the input parameters include water supply temperature, return water temperature difference, ambient temperature and humidity, and historical energy consumption trends; cascading the operating condition division model and the prediction network to form a hybrid prediction architecture for the host energy consumption.

[0136] The construction of the variable frequency water pump model includes: fitting a cubic relationship at a rated frequency reference point to obtain a theoretical flow-frequency curve; collecting actual pipe network characteristic data to calculate the flow attenuation coefficient at each frequency point; and constructing a modified cubic equation including a pipe network damping coefficient.

[0137] A differential equation model of heat transfer for the surface cooler is established, and a return water temperature data-driven model based on Att-BiGRU is constructed. The input features include two-order lagged outdoor temperature and humidity, indoor temperature, and water supply parameters.

[0138] Design a model fusion strategy to enable the data-driven model to perform compensation prediction when the physical model calculation residual exceeds the threshold.

[0139] The construction of the software-in-the-loop simulation platform specifically includes: constructing a building thermodynamic equation module in Matlab / Simulink and integrating the EnergyPlus core calculation engine; establishing a real-time data channel between KepserverEX and the simulation model through the OPC UA protocol; embedding the control logic code library of the actual equipment in the simulation environment to achieve lossless migration of the control algorithm.

[0140] The present invention also discloses the following contents to help understand the technical solution.

[0141] The realization of full-process automation, informatization, digitization and advanced control technology of air-conditioning systems often cannot be separated from software that integrates relevant intelligent algorithms. Traditional software testing methods usually rely on actual systems and hardware, such as hardware-in-the-loop simulators. For the development of early software and related control algorithms, it is almost impractical to use actual systems for testing. Especially for large public buildings, it is difficult to ensure the stable operation of the system during testing due to frequent control of equipment. Therefore, the actual test time is generally short and discontinuous. In order to reduce the cost and risk of testing, a software-in-the-loop (SIL) simulation platform for air-conditioning control systems was established to test the source code before the system and hardware configuration, so as to verify the effectiveness of the proposed predictive control algorithm for air-conditioning systems at an early stage.

[0142] Simulation platform construction: Communication methods between various sensors and equipment in the station and the software platform Figure 1As shown in the figure. Sensor and device signals are uploaded to the automated device data connection and acquisition software KepserverEX via wired or wireless networks, edge nodes, etc. After converting the signals into common data forms, the KepserverEX software uploads them to the software platform via communication protocols such as Modbus and OPC. After processing and calculating the collected data using intelligent algorithms, the software platform issues control signals and sends control instructions to the devices via the KepserverEX software. To fully simulate the communication method between the station sensors and devices and the software platform, a station air-conditioning control system simulation platform as shown in Figure 2 is built. This simulation platform uses Matlab software to establish a building data model and controller algorithm, and uses KepserverEX software to achieve the communication connection between the building data model and the controller. In previous studies, software such as Energyplus and Trnsys were often used to simulate the building environment, but the simulated building models may have a large gap from the actual project. Therefore, a digital building and equipment model is established using the actually collected station environment and equipment data to simulate the real building and equipment status, and at the same time, it can also transmit the simulation signals of relevant internal and external disturbances established from real data. After the MPC controller test is completed, it can be directly used for the control of the actual system, or it can be integrated with other software platforms for the intelligent optimization control of the air-conditioning system.

[0143] Establishment of building data model

[0144] A digital virtual indoor temperature and equipment simulation model is established through the actually collected station environment and equipment data to simulate the real indoor and equipment status.

[0145] (1) Virtual indoor temperature simulation model

[0146] A virtual indoor temperature simulation model is constructed based on the digital twin modeling method of air-conditioning load with time-shift correlation of characteristic variables and deep learning. First, the outdoor dry-bulb temperature (T o ), relative humidity (H o ), wind speed (W f ), wind direction (W d ), solar radiation intensity (R o ) and indoor occupants (N p ) are used as initial environmental impact factors for correlation analysis. The MIC values of each impact factor and the indoor temperature are as shown in Figure 3 .

[0147] From Figure 3 , it can be seen that the indoor temperature has a relatively high correlation with T o with a lag step of 0-3, H o with a lag step of 0-2, and N p with a lag step of 0, and has a relatively low correlation with Wf , W d and R o have a low correlation. T o , H o and N p have maximum MIC values of 0.70, 0.63, and 0.64 respectively. The PCA algorithm is used to perform dimensionality reduction on T o and H o . The results of different dimensional parameters are shown in Tables 1 and 2. Among them, the cumulative variance explanation rate of the first two principal components of T o is 98.35%, and the cumulative variance explanation rate of the first two principal components of H o is 99.08%. Therefore, two principal components are selected as the input variables related to T o , denoted as T oa1 and T oa2 respectively; two principal components are selected as the input variables related to H o , denoted as H oa1 and H oa2 respectively; the number of people indoors with a lag step of 0 is denoted as N p .

[0148] Table 1 PCA dimensionality reduction results of outdoor temperature (T o )

[0149]

[0150] Table 2 PCA dimensionality reduction results of outdoor relative humidity (H o )

[0151]

[0152] The water supply temperatures T sup0 , T sup1 with a lag step of 0 - 1, the water supply flow rates m sup0 , m sup1 with a lag step of 0 - 1, and the indoor temperature T n at the prediction moment are also used as input variables of the prediction model for subsequent modeling. The indoor temperature prediction model includes T oa1 , T oa2 , H oa1 , H oa2 , N p , T n , T sup0 , T sup1 , m sup0 and m sup1 a total of 10 input variables. The above variables represent variables related to outdoor dry - bulb temperature, relative humidity, number of people indoors, indoor temperature, water supply temperature, and water supply flow rate respectively.

[0153] Four algorithms, namely Att-LSTM, Att-BiLSTM, Att-GRU, and Att-BiGRU, are used to construct a virtual indoor temperature simulation model. The simulation accuracy on the model validation set is shown in Table 3.

[0154] Table 3 Comparison of prediction performance of different algorithms

[0155]

[0156] As can be seen from Table 3, the Att-BiGRU algorithm has better simulation accuracy on the model validation set than the other three algorithms. Its MAE, MAPE, and RMSE values are 0.069, 0.002, and 0.112 respectively. Thus, it can be seen that the virtual indoor temperature simulation model established by the Att-BiGRU model has extremely high accuracy and can be used to simulate the real building environment.

[0157] (2) Virtual air conditioning equipment model

[0158] Air source heat pump model: An air source heat pump mainly consists of components such as a compressor, an evaporator, a condenser, and a throttling device. Taking the commonly used compression-type air source heat pump as an example, its working principle is as follows: First, the compressor does work on the refrigerant to change the refrigerant vapor in the low-temperature and low-pressure state into a high-temperature and high-pressure state and discharge it into the condenser, where it is converted into liquid refrigerant through condensation and heat release. After passing through the throttling device, the pressure and temperature of the liquid refrigerant decrease simultaneously, and it is converted into a gas-liquid two-phase state and enters the evaporator, where it absorbs heat from the environmental medium and is converted into refrigerant gas and enters the compressor, and the above process is continuously repeated in the system to complete heat transfer, achieving the purpose of refrigeration or heating. We established the main engine energy consumption model of the air source heat pump, and the establishment process is as follows:

[0159] The calculation formula for the coefficient of performance COP of the heat pump is shown in Equation (4-1):

[0160]

[0161] In the formula: T low is the temperature of the low-temperature heat source, which is the outlet water temperature of the heat pump unit under the refrigeration condition, °C; T high is the temperature of the high-temperature heat source, which is the outdoor ambient temperature under the refrigeration condition, °C.

[0162] In addition, we also established a COP prediction model for the air source heat pump unit based on fuzzy clustering and a stacked wide learning system.

[0163] The main engine energy consumption W ASHP The calculation formula is shown in Equation (4-2):

[0164] W ASHP = Q ASHP / COP*ΔT (4-2)

[0165] Where: Q ASHP is the system refrigerating capacity per unit time, in kW; ΔT is the system operation time, in h.

[0166] It can be seen from Equation (6-2) that the host energy consumption is related to the system refrigerating capacity and COP. The system energy consumption is related to the supply water temperature T sup , the return water temperature T ret , the flow rate m sup , the outdoor ambient temperature T o , the relative humidity H o , the number of indoor people N p , the indoor temperature T N and the historical cooling capacity Q his . Therefore, the host energy consumption model can be expressed as shown in Equation (4-3):

[0167] W ASHP = f ASHP (T sup , T ret , m sup , T o , H o , N p , T N , Q his )*ΔT (4-3)

[0168] Where: f ASHP is a data-driven model related to the energy consumption calculation of the air source heat pump unit.

[0169] Pump model

[0170] For a variable-frequency pump, according to the similarity law, when the fluid flow process simultaneously satisfies geometric similarity, kinematic similarity, and dynamic similarity, the flow rate of the pump is proportional to the frequency, and the electric power is proportional to the cube of the frequency. The relationships between the pump flow rate m and the electric power Q and the pump frequency n are shown in Equations (4-4) and (4-5):

[0171]

[0172] Where: m1 and m0 are the flow rates of the pump under the actual working condition and the rated working condition respectively, in m 3 / h; n1 and n0 are the frequencies of the pump under the actual working condition and the rated working condition respectively, in Hz; Q1 and Q0 are the electric powers of the pump under the actual working condition and the rated working condition respectively, in kW.

[0173] For a water system with an actively adjustable water flow rate for the terminal load, the relationships between the flow rate and the electric power of the pump under the actual working condition and the frequency can be expressed by Equations (4-6) and (4-7).

[0174]

[0175] Where: a, b, c, d, e, and f are fitting coefficients.

[0176] Calculated according to the collected data, the relationships between the flow rate and electric power of the water pump under actual working conditions and the frequency are shown in Equations (4-8) and (4-9).

[0177] m pump = 6.5001n pump -40.5961 (4-8)

[0178] Q pump = -0.0083n pump 3 +1.0829n pump 2 -45.6339n pump +641.4388 (4-9)

[0179] Where: m pump is the flow rate of the water pump in the actual working condition of this engineering case, m 3 / h; n pump is the frequency of the water pump in the actual working condition, Hz; Q pump is the electric power of the water pump in the actual working condition, kW.

[0180] The energy consumption W of the water pump pump is calculated as shown in Equation (4-10):

[0181] W pump = Q pump *ΔT (4-10)

[0182] Combined air handling unit model

[0183] The simulation model of the combined air handling unit established by the present invention includes a return water temperature model and a fan energy consumption model.

[0184] A. Return water temperature model

[0185] The combined air handling unit mainly consists of equipment such as a surface cooler, a humidifier, a heater, and a fan. During the refrigeration season, the surface cooler is mainly used to cool and dehumidify the air. For a primary return constant air volume air conditioning system, the mixed indoor air and outdoor fresh air can obtain the initial state dry bulb temperature of the air treatment as t1, the enthalpy value as h1, and the wet bulb temperature as t s1 , the air volume to be treated is G, the chilled water flow rate is m, and the initial temperature of the chilled water is T sup , then the final state t2 and the return water temperature T that the surface cooler can reach for treating the air retThe calculation process is as follows:

[0186] (1) Determine the face velocity V of the finned tube heat exchanger y and the water velocity w. The calculation formulas are shown in Formulas (4-11) and (4-12):

[0187]

[0188] In the formula: F y is the frontal area of the finned tube heat exchanger, m 2 ; ρ is the density of water, kg / m 3 ; f w is the cross-sectional area of water flow, m 2 .

[0189] (2) Determine the general heat transfer efficiency E' provided by the finned tube heat exchanger. This value can be obtained from the equipment sample.

[0190] (3) Assume that the dry bulb temperature of the air at the final state is t2.

[0191] Generally, the dry bulb temperature of the air at the final state can be determined by t2 = t w1 +(4 - 6)°C. The wet bulb temperature t s2 at the final state is calculated as shown in Formula (4-13):

[0192] t s2 = t2 - (t1 - t s1 )(1 - E') (4-13)

[0193] The enthalpy value h2 at the final state is calculated as shown in Formula (4-14):

[0194] h2 = 0.0707t s 2 2 + 0.6452t s2 + 16.18 (4-14)

[0195] (4) Determine the dehumidification coefficient ξ. The calculation formula is shown in (4-15):

[0196]

[0197] Among them, c p is the specific heat capacity at constant pressure of water, taking 4.2×10 3 J / (kg·°C);

[0198] (5) Determine the heat transfer coefficient K s .

[0199] The empirical formula for the heat transfer coefficient of the heat exchanger in the dehumidifying and cooling process is shown in Formula (4-16):

[0200]

[0201] In the formula: A, P, B, m, and n are coefficients and exponents obtained from experiments.

[0202] (6) Calculate the total heat exchange efficiency E' that the surface cooler can achieve g , and the calculation formula is as shown in formulas (4-17)-(4-19):

[0203]

[0204] In the formula, F represents the frontal area of the surface cooler;

[0205]

[0206] In the formula, c represents the specific heat capacity of water;

[0207] (7) Calculate the required total heat exchange efficiency E g and compare it with E' g , and the calculation formula of E g is as shown in formula (4-20):

[0208]

[0209] When |E g -E' g | ≤ δ (δ is the allowable error, generally taken as 0.01), it is proved that the assumed t2 is appropriate. If |E g -E' g | > δ, then t2 should be reset and recalculated.

[0210] (8) Calculate the return water temperature T of the chilled water ret , and the calculation formula is as shown in formula (4-21).

[0211]

[0212] From the above analysis, it can be seen that for the air handling unit, it is advisable to establish a black-box model of the unit in a data-driven manner. In this embodiment, a digital twin modeling method based on the time-shift correlation of characteristic variables and deep learning is used to establish a prediction model for the return water temperature after heat exchange. The modeling process is as follows:

[0213] (1) Determine the input variables

[0214] Through the analysis of the above-mentioned return water temperature calculation process, it can be obtained that the return water temperature is related to the outdoor dry-bulb temperature, outdoor wet-bulb temperature, return air dry-bulb temperature, return air wet-bulb temperature, supply water temperature and supply water flow rate. The outdoor dry-bulb temperature can be replaced by the outdoor temperature, and the return air dry-bulb temperature can be replaced by the indoor temperature. Since the wet-bulb temperature is difficult to measure, the relative humidity is used to replace the wet-bulb temperature as the input variable. The initial influencing factors of the return water temperature prediction model can be determined as the outdoor temperature T o , outdoor relative humidity H o , indoor dry-bulb temperature T n , indoor relative humidity H n , supply water temperature T sup and supply water flow rate m sup . The MIC values of each influencing factor and the return water temperature are as Figure 4 shown.

[0215] From Figure 4 it can be seen that the return water temperature is related to T with a lag step of 0-1 o , H with a lag step of 0-1 o , T with a lag step of 0-1 n , T with a lag step of 0-1 sup and m with a lag step of 0-1 sup . The input variables are respectively denoted as T o1 , T o2 , H o1 , H o2 , T n1 , T n2 , T sup1 , T sup2 , m sup1 and m sup2 , and the above variables respectively represent the variables related to the outdoor dry-bulb temperature, outdoor relative humidity, indoor dry-bulb temperature, supply water temperature and supply water flow rate.

[0216] (2) Model establishment

[0217] Four algorithms, Att-LSTM, Att-BiLSTM, Att-GRU, and Att-BiGRU, are used to construct a simulation model of the return water temperature. The simulation accuracy on the model validation set is shown in Table 4.

[0218] It can be seen from Table 4 that the simulation accuracy of the Att-BiGRU algorithm on the model validation set is better than that of the other three algorithms, and its MAE, MAPE, and RMSE values are 0.102, 0.006, and 0.189 respectively. Thus, it can be seen that the simulation model of the return water temperature established by the Att-BiGRU model has extremely high accuracy and can be used to simulate the real return water temperature.

[0219] Table 4 Comparison of prediction performance of different algorithms

[0220]

[0221] B. Fan energy consumption model

[0222] The energy consumption of the modular air handling unit is the fan energy consumption. According to the similarity law, the power of the fan is proportional to the cube of its operating frequency, i.e.:

[0223]

[0224] Where: n fan1 and n fan2 are the frequencies of the fan under actual and rated conditions respectively, Hz; Q fan1 and Q fan0 are the electric powers of the fan under actual and rated conditions respectively, kW.

[0225] The fan in the modular air handling unit is a constant-frequency fan, and its operating air volume and power remain unchanged. Therefore, the calculation formula for the fan energy consumption W fan is shown in Equation (4-23):

[0226] W fan = Q fan *ΔT (4-23)

[0227] Where: Q fan is the actual operating power of the fan, kW.

[0228] Through Equation (6-3), Equation (6-10) and Equation (6-23), the calculation formula for the total energy consumption of the air conditioning system can be obtained as shown in Equation (4-24):

[0229] W total = (f ASHP + Q pump + Q fan )*ΔT (4-24)

[0230] Although the preferred embodiments of the present invention have been described in detail, those skilled in the art can still make other changes and modifications to these embodiments after understanding their basic creative concepts. Therefore, the appended claims are intended to cover these preferred embodiments and all changes and modifications that fall within the scope of the present invention. The above description is only the preferred embodiment of the present invention and is not intended to limit its scope. It should be clear that any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-control object modeling method for air conditioning system, characterized in that: include: Acquire multi-source operation data collected by air-conditioning system sensors, the multi-source operation data including outdoor temperature To, outdoor relative humidity Ho, indoor temperature Tn, water supply temperature Tsup, return water temperature Tret, water supply flow msup, water pump frequency npump and indoor occupancy number Np; perform principal component analysis (PCA) and time-shift correlation analysis on the multi-source operation data to screen out core input variables corresponding to the target control object; wherein, The target control objects include virtual indoor temperature, energy consumption of air source heat pump unit, water pump energy consumption and fan energy consumption; The time-shift correlation analysis determines the lag step of each variable through the maximum information coefficient (MIC) to screen the key time series characteristics; based on the screened core input variables, the data-driven model of each control object is constructed respectively, and the model is trained and verified to form a combined multi-control object simulation framework; wherein, the combined multi-control object simulation framework is used to predict the energy consumption and environmental status of the air-conditioning system.

2. The multi-control object modeling method for air conditioning system according to claim 1 is characterized in that: The steps of constructing a virtual indoor temperature model include: determining input variables through MIC analysis, including the 0-3 step lagged characteristics of the outdoor temperature To, the 0-2 step lagged characteristics of the outdoor humidity Ho and the current number of people Np indoors; performing PCA dimensionality reduction processing on the outdoor temperature and humidity, and taking the principal components with cumulative variance contribution rates ≥ 98% as the input variables after dimensionality reduction, which are recorded as Toa1, Toa2 and Hoa1, Hoa2 respectively; combining the reduced variables with the historical water supply temperature (Tsup0-Tsup1), the historical water supply flow (msup0-msup1) and the current indoor temperature (Tn) to construct a training set with 10 input variables; using the Att-BiGRU algorithm to model the training set, optimizing the network parameters through back propagation, and obtaining a virtual indoor temperature simulation model with a prediction error MAE≤0.07 and RMSE≤0.

12.

3. The multi-control object modeling method for air conditioning system according to claim 1, characterized in that: The steps of constructing the energy consumption model of the air source heat pump unit include: screening input variables, including the current supply water temperature Tsup, return water temperature Tret, flow rate msup, outdoor temperature To, outdoor humidity Ho, indoor personnel Np, indoor temperature Tn and historical cooling capacity Qhis; the input variables are implanted into the stacked width learning system to establish an energy consumption prediction sub-model, whose output is the heat pump performance coefficient COP, and the calculation relationship satisfies: COP = f(T low ,T high ), where T low is the outlet water temperature, T high is the outdoor temperature; ASHP =Q ASHP / COP*ΔT outputs the energy consumption value of the output host, where Q is the system cooling capacity per unit time, kW; ΔT is the system operating time, h.

4. The multi-control object modeling method for air conditioning system according to claim 1, characterized in that: When building a water pump energy consumption model, do the following: The flow rate of the water pump is proportional to the frequency, and the electric power is proportional to the cube of the frequency; the relationship between the water pump flow rate m, the electric power Q and the water pump frequency n is shown in equations (1-1) and (1-2): Where: m1 and m0 are the flow rates of the pump under actual working conditions and rated working conditions respectively, m 3 / h; n1 and n0 are the frequencies of the water pump under actual working conditions and rated working conditions, Hz; Q1 and Q0 are the electrical powers of the water pump under actual working conditions and rated working conditions, kW; The relationship between the flow rate, electric power and frequency of the water pump under actual working conditions can be expressed by equations (1-3) and (1-4); Where: a, b, c, d, e, f are the coefficients fitted by measured data.

5. The multi-control object modeling method for air conditioning system according to claim 1, characterized in that: Construct a combined air conditioning unit model, including a return water temperature prediction model: For a single return constant air volume air conditioning system, the indoor air is mixed with the outdoor fresh air to obtain the initial state of air treatment with a dry bulb temperature of t1, an enthalpy value of h1, and a wet bulb temperature of t s1 , the amount of air to be processed is G, the chilled water flow is m, and the initial temperature of the chilled water is T sup , then the final state t2 and return water temperature T that the surface cooler can reach when processing air ret The process is as follows: A1. Calculate the wind speed V facing the surface cooler y And water flow rate w, the calculation formula is shown in formula (2-1) and formula (2-2): Where: F y is the windward area of ​​the surface cooler, m 2 ; ρ is the density of water, kg / m 3 ;f w is the cross-sectional area through which water can pass, m 2 ; A2. Calculate the general heat exchange efficiency E' that the surface cooler can provide. This value can be obtained from the equipment sample; A3. Assume that the dry-bulb temperature of the air in the final state is t2; According to t2=t w1 +(4~6)℃ determines the final dry bulb temperature of the air; the final wet bulb temperature t s2 The calculation formula is shown in formula (2-3): t s2 =t2-(t1-t s1 )(1-E') (2-3) The calculation formula of the final state enthalpy h2 is shown in formula (2-4): A4. Calculate the moisture coefficient ξ, as shown in (2-5): A5. Calculate the heat transfer coefficient K s ; The empirical formula of heat transfer coefficient of heat exchanger dehumidification cooling process is shown in formula (2-6): Where: A, P, B, m, n are coefficients and exponents obtained from experiments; A6. Calculate the total heat exchange efficiency E' that the surface cooler can achieve g , the calculation formula is shown in formula (2-7) to formula (2-9): A7. Calculate the required total heat exchange efficiency E g And with E' g In comparison, E g The calculation formula is shown in formula (2-10): When|E g -E' g |≤δ (usually 0.01), it proves that the assumed t2 is appropriate; if |E g -E' g |>δ, t2 should be reset and the calculation should be repeated; A8. Calculate the chilled water return temperature T ret , the calculation formula is shown in formula (2-11); 6. The multi-control object modeling method for air conditioning system according to claim 1, characterized in that: Construct a combined air conditioning unit model, including: establish a return water temperature prediction model: through MIC analysis, determine the input variables as To lag 1 step, Ho lag 1 step, Tn or Tn lag 1 step, Tsup lag 1 step and msup lag 1 step, and use the Att-BiGRU algorithm to train to obtain a return water temperature prediction sub-model with MAE ≤ 0.11; establish a fixed-frequency fan energy consumption model: the energy consumption of the combined air conditioning unit is the fan energy consumption. According to the similarity law, the power of the fan is proportional to the cube of its operating frequency. The fan in the combined air conditioning unit is a fixed-frequency fan, and its operating air volume and power remain unchanged; the fan energy consumption W fan The calculation formula is shown in formula (3-1): W fan =Q fan *ΔT (3-1) Where: Q fan is the actual operating power of the fan, kW.

7. The multi-control object modeling method for air conditioning system according to claim 6, characterized in that: When establishing the return water temperature prediction model, the model training and verification steps include: dividing the sample data into training set and verification set at a ratio of 7:3, using the first 10 time steps to predict the next time step output; using the early stopping method to monitor the verification set error, and terminating the training when the RMSE drops by less than 1% for 10 consecutive iterations; calculating the mean absolute error, mean absolute percentage error, and root mean square error of the verification set, selecting Att-BiGRU as the optimal basic algorithm, and optimizing its model weights by gradient descent.

8. The multi-control object modeling method for air conditioning system according to claim 1, characterized in that: It also includes a model integration method based on the simulation platform: building a building data model in the Matlab environment, establishing a Modbus / OPC communication link with the MPC controller through the kepserverex software; integrating the virtual indoor temperature model, air source heat pump model, water pump model and air conditioning unit model in parallel, and ensuring the consistency of simulation steps through the time synchronization module; generating model output signals including: real-time temperature prediction value Tn+1, host energy consumption W ASHP , Water pump energy consumption W pump , fan energy consumption W fan , and map it to the register address of kepserverex for the controller to read.

9. The multi-control object modeling method for air conditioning system according to claim 8, characterized in that: The communication link construction specifically includes: creating a virtual device in kepserverex, binding the input and output variables of the Matlab model to the Modbus TCP protocol registers; setting the simulation step to 5 minutes, and completing the closed-loop process of data writing → model calculation → register update → controller reading within each step; configuring the OPC UA client on the controller side, and regularly reading the predicted temperature and energy consumption data in the register.

10. The multi-control object modeling method for air conditioning system according to claim 1, characterized in that: It also includes a model online update method: deploy an operation data monitoring module to collect the actual temperature value T_real and the model prediction value T_pred in real time; when the daily average prediction error exceeds the threshold e, the model retraining process is triggered; using incremental learning, the latest 7 days of data is added to the training set, the Att-BiGRU network weights are updated, and the model prediction accuracy is maintained.

Images