Prediction method for refrigeration load of industrial clean workshop

Abstract

The invention relates to a method for predicting the refrigeration load of an industrial clean workshop, and the method comprises the steps: building a heating ventilation air-conditioning system model which comprises a room heat model, a fresh air processing unit model, a dry cooling coil model and a fan filtering unit model, so as to build an interaction relation between an air side and the room model; selecting a sensitive input parameter set influencing the refrigeration load output by the heating ventilation air-conditioning system model, obtaining missing input parameters by adopting a PID (Proportion Integration Differentiation) controller and a feedback backstepping mode, and predicting to obtain missing input parameters in any time period based on a machine learning model and training set data; after the optimal input parameter combination selected based on genetic algorithm calculation processing is input into the HVAC system model, and the cooling load output by the HVAC system model is used as the predicted value of the cooling load of the industrial clean plant, efficient and accurate modeling of the demand side HVAC system is realized, and the real characteristics of the HVAC system are captured more accurately. And the prediction precision of the refrigeration load of the industrial clean workshop is improved.

Description

Technical Field

[0001] The invention relates to the field of industrial clean workshops, and in particular to a method for predicting refrigeration load of industrial clean workshops. Background Art

[0002] Industrial cleanrooms have been widely used in high-tech manufacturing industries such as semiconductor manufacturing, LED production, and aerospace, which put forward more stringent requirements for indoor environmental control such as temperature, humidity, and particle concentration. For this reason, the energy consumption of industrial cleanrooms is much higher than that of ordinary commercial buildings or office buildings. Among them, the HVAC system is one of the subsystems with the highest energy consumption in industrial cleanrooms, accounting for more than 50% of the total energy consumption of industrial cleanrooms. Therefore, reducing the energy consumption of HVAC systems in industrial cleanrooms is crucial to achieving the energy conservation and emission reduction goals of the industrial sector.

[0003] Existing industrial cleanrooms usually use regional cooling systems, which are usually composed of three parts: water-side system, air-side system and production workshop. The water-side system consumes energy to produce chilled water, and then the air-side system uses the chilled water produced by the water-side system to cool and dehumidify the production workshop. In this complex refrigeration system, the water-side system and the air-side system jointly maintain the stability of the production workshop environment through a series of integrated feedback controllers.

[0004] In order to reduce the energy consumption of HVAC systems in industrial clean plants, the current approach is to model the HVAC systems on the demand side, and then optimize the control configuration of the water side system such as chilled water flow, water supply temperature and equipment sequencing to complete the prediction of the refrigeration load of the industrial clean plant, thereby achieving the goal of reducing energy consumption.

[0005] However, the existing prediction methods for the cooling load of industrial clean plants have shortcomings: when constructing the HVAC system model on the demand side, only the air-side model or the room model is considered as independent models, but the complex interaction between the air-side model and the room model is not considered. It is impossible to construct a HVAC system model that better reflects the real characteristics, and it is impossible to make an accurate prediction of the cooling load of the industrial clean plant. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method for predicting the refrigeration load of an industrial clean room, which fully considers the interaction between the air side model and the room model to improve the prediction accuracy, with respect to the above-mentioned prior art.

[0007] The technical solution adopted by the present invention to solve the above technical problem is: a method for predicting the refrigeration load of an industrial clean room, characterized in that it comprises the following steps:

[0008] Step 1, constructing a HVAC system model including a room thermal model, a fresh air handling unit model, a dry cooling coil model, and a fan filter unit model;

[0009] Step 2, performing a global sensitivity analysis on each input parameter of the constructed HVAC system model, respectively calculating the influence value of each input parameter on the cooling load output by the HVAC system model, and selecting multiple input parameters with an important influence degree on the cooling load output by the HVAC system model according to all the influence values ​​obtained, so as to form a sensitive input parameter set with the selected multiple input parameters; wherein the influence value of the input parameter on the cooling load output by the HVAC system model is positively correlated with its important influence degree;

[0010] Step 3: Based on the HVAC system model, the actual measurement data of cooling load in the preset historical period obtained in advance is used, and a PID controller is used to perform feedback adjustment and back-stepping on the missing input parameters in the sensitive input parameter set to obtain the missing input parameters obtained after back-stepping, and the missing input parameters obtained after back-stepping and all other input parameters in the sensitive input parameter set together form the training set data;

[0011] Step 4: Input the formed training set data and the sensitive input parameter set in any time period into the machine learning model for prediction processing, and predict the missing input parameters in the arbitrary time period;

[0012] Step 5, using the remaining input parameters in the sensitive input parameter set and excluding the missing input parameter obtained by prediction to form a plurality of non-completely identical input parameter combinations, and inputting each input parameter combination into the HVAC system model respectively, and obtaining the HVAC system model output value corresponding to each input parameter combination respectively;

[0013] Step 6, using a genetic algorithm tool to calculate the output error values ​​between the output values ​​of each HVAC system model and the actual measurement data of the cooling load in the preset historical period, and selecting the input parameter combination corresponding to the output error value with the minimum error value as the optimal input parameter combination of the HVAC system model;

[0014] Step 7: input the actual value of the input parameter corresponding to each input parameter in the optimal input parameter combination into the HVAC system model, and use the cooling load output by the HVAC system model as the predicted value of the refrigeration load of the industrial clean room.

[0015] Improved, in the method for predicting the refrigeration load of the industrial clean room, in step 1:

[0016] The room thermal model uses a capacitor-resistance model to describe the dynamic change of indoor temperature. The transient energy balance equation of the room thermal model is as follows:

[0017]

[0018] Among them, C room is the heat capacity of the room, T room is the room temperature, T ext is the outdoor temperature of the room, R th is the thermal resistance of the room, Q int is the impact of the room’s internal heat load on the indoor temperature, represents the rate of change of the room's indoor temperature over time, Indicates the transient change value of the heat in the room. It indicates the influence of external heat conducted through thermal resistance on the indoor temperature of the room;

[0019] The fresh air handling unit model includes a wet cooling coil model, a reheat coil model and a fan model. The fresh air handling unit model processes the sensible heat and latent heat in the air. The dry cooling coil model and the fan filter unit model together form a fan coil unit. The transient energy balance equation corresponding to the fan coil unit is as follows:

[0020]

[0021] in, is the mass flow rate of chilled water in the dry cooling coil, c pw is the specific heat capacity of the chilled water in the dry cooling coil, T cw,in is the inlet temperature of the chilled water at the inlet of the dry cooling coil, T cw,out is the outlet temperature of the chilled water at the outlet of the dry cooling coil, The heat transferred by the chilled water in the dry cooling coil; is the air mass flow rate in the fresh air handling unit model, c pa is the specific heat capacity of air, T air,in Indicates the air temperature before passing through the dry cooling coil, T air,out Indicates the air temperature after passing through the dry cooling coil;

[0022] The transient energy balance equation of the heat transfer process corresponding to the dry cooling coil model is as follows:

[0023]

[0024] The transient energy balance equation of the wet cooling coil model of the fresh air handling unit model is as follows:

[0025]

[0026] h=c pa ·T air +ω·h fg ;

[0027] h in =c pa ·T in +ω in ·h fg ;

[0028] h out =c pa ·T out +ω out ·h fg ;

[0029] Among them, h in is the enthalpy of the air entering the wet cooling coil, h out is the enthalpy of the air leaving the wet cooling coil, h in -h out Contains the changes in sensible heat and latent heat in the wet cooling coil model, h is the enthalpy of air, T air is the air temperature, ω is the air humidity ratio, h fg is the latent heat of vaporization of water; T in is the air temperature entering the wet cooling coil, ω in is the humidity ratio of the air entering the wet cooling coil, T out is the air temperature leaving the wet cooling coil, ω out is the humidity ratio of the air leaving the wet cooling coil.

[0030] Furthermore, in the method for predicting the refrigeration load of the industrial clean room, in step 2, the sensitivity value of the refrigeration load output by the HVAC system model to any input parameter is expressed as follows:

[0031]

[0032] θ=(θ1,θ2,…,θ n );θ'=(θ1,θ2,…,θ i-1 ,θ i +△,θ i+1 ,…,θ n );

[0033]

[0034] Among them, d(θ i ) represents the cooling load output by the HVAC system model for the i-th input parameter θ i The sensitivity value, θ iis the i-th input parameter in the HVAC system model that affects the cooling load output by the HVAC system model, n is the total number of input parameters in the HVAC system model that affect the cooling load output by the HVAC system model, △ is the step size for perturbing the input parameters of the HVAC system model in the global sensitivity analysis, θ is the benchmark parameter set, θ' is the i-th input parameter θ in the benchmark parameter set θ i The new parameter set after being disturbed by a step length △; f(θ) represents the cooling load corresponding to the output when the input parameter set of the HVAC system model is the benchmark parameter set θ; f(θ') represents the cooling load corresponding to the output when the input parameter set of the HVAC system model is the new parameter set θ'; H represents the level of detail after each input parameter of the HVAC system model is discretized in the global sensitivity analysis;

[0035] Input parameter θ i The impact value on the cooling load output by the HVAC system model is marked as

[0036]

[0037] in, is the input parameter θ i The average effect value of the cooling load output by the HVAC system model, r is the total number of trajectories selected for global sensitivity analysis, d ij (θ) represents the jth trajectory among the r trajectories for the input parameter θ i The perturbation effect value of the basic effect, σ i is the input parameter θ i The standard deviation of the effect on the cooling load output by the HVAC system model.

[0038] Further improved, in the prediction method of the refrigeration load of the industrial clean room, in step 3, the control formula of the PID controller is as follows:

[0039]

[0040] Where IL(t) is the missing input parameter obtained after inference, e(t) is the error between the observed cooling load and the cooling load output by the constructed HVAC system model, and K p is the proportional gain, K i is the integral gain in the PID controller.

[0041] Furthermore, in the prediction method of the refrigeration load of the industrial clean room, in step 4, the machine learning model is an XGBoos model, and the XGBoos model is used to perform machine learning inference on the input parameters that are missing and are in the sensitive input parameter set. The objective function of the XGBoos model is expressed as follows:

[0042]

[0043] Among them, L(φ) is the objective function of the XGBoost model, Represents the true value y of the vth sample v and the predicted value of the vth sample The loss function of the error between, K represents the total number of decision regression trees that form the XGBoost training model, and f k represents the kth decision regression tree among K decision regression trees, Ω(f k ) is the model complexity penalty term corresponding to the kth decision regression tree.

[0044] Compared with the prior art, the advantages of the present invention are:

[0045] First, the prediction method of the refrigeration load of the industrial clean room of the invention constructs a HVAC system model including a room thermal model, a fresh air handling unit model, a dry cooling coil model and a fan filter unit model, establishes the interaction relationship between the air side and the room model, and uses the global sensitivity analysis method to select multiple input parameter sets that deeply affect the cold load output by the HVAC system model to form a sensitive input parameter set, and uses a PID controller and feedback backstepping method to obtain the missing input parameters and further form training set data, and predicts the missing input parameters in any time period based on the machine learning model, and then forms different input parameter combinations to input into the HVAC system model, and selects the optimal input parameter combination based on the genetic algorithm calculation and processing, and after the optimal input parameter combination is input into the HVAC system model, the cold load corresponding to the output of the HVAC system model is used as the predicted value of the refrigeration load of the industrial clean room. In this way, the constructed HVAC system model fully considers the interaction between the air side model and the room model, and efficiently and accurately models the demand side HVAC system, more accurately captures the real characteristics of the HVAC system, and improves the prediction accuracy of the refrigeration load of the industrial clean room.

[0046] Secondly, the machine learning model is used to process the input training set data and the sensitive input parameter set within any time period to predict the missing input parameters within the arbitrary time period. This overcomes the limitations of traditional modeling methods in terms of data missing and system complexity, avoids data missing, and improves the robustness of the model and the accuracy of refrigeration load prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic flow chart of a method for predicting refrigeration load of an industrial clean room in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the refrigeration load prediction effect corresponding to the prediction method of the refrigeration load of the traditional industrial clean room;

[0049] Figure 3 It is a schematic diagram of the refrigeration load prediction effect corresponding to the prediction method of the refrigeration load of the industrial clean room in the embodiment of the present invention. DETAILED DESCRIPTION

[0050] The present invention is further described in detail below with reference to the accompanying drawings.

[0051] This embodiment provides a method for predicting the refrigeration load of an industrial clean plant. Specifically, this embodiment takes a large clean room production workshop in a high-tech manufacturing industrial park in Sichuan Province as an example. The industrial park includes a clean production workshop with an area of ​​approximately 23,800 square meters. A large number of Internet of Things (IoT) sensor devices are installed in the workshop to collect multi-dimensional environmental and system operation data in real time, including but not limited to indoor and outdoor temperature, humidity, chilled water inlet and outlet temperature, and air supply temperature of the air unit. All data is transmitted through the Internet of Things and saved to a dedicated background data platform, which effectively ensures the integrity and timeliness of the data. Sensor data from May 25 to July 7, 2023 (with a time resolution of 15 minutes) was used as a test for the proposed method. See Figure 1 As shown, the method for predicting the refrigeration load of an industrial clean room in this embodiment includes the following steps 1 to 7:

[0052] Step 1, constructing a HVAC system model including a room thermal model, a fresh air handling unit model, a dry cooling coil model, and a fan filter unit model;

[0053] Specifically, in this embodiment, the room thermal model here adopts a capacitor-resistance model to describe the dynamic change of the indoor temperature. The transient energy balance equation of the room thermal model is as follows:

[0054]

[0055] Among them, C room is the heat capacity of the room, T room is the room temperature, T ext is the outdoor temperature of the room, R th is the thermal resistance of the room, Q int is the impact of the room’s internal heat load on the indoor temperature, represents the rate of change of the room's indoor temperature over time, Indicates the transient change value of the heat in the room. It represents the influence of external heat conducted through thermal resistance on the indoor temperature of the room. Based on the transient energy balance equation of the room thermal model, it can simulate the transient temperature response in the room and further link with the air-side system model to form a complete parameterized dynamic model of the clean room demand-side HVAC system, which fully considers the interaction between the room model and the air-side system.

[0056] The fresh air handling unit model includes a wet cooling coil model, a reheat coil model and a fan model. The fresh air handling unit model processes the sensible heat and latent heat in the air. The dry cooling coil model and the fan filter unit model together form a fan coil unit. The transient energy balance equation corresponding to the fan coil unit is as follows:

[0057]

[0058]

[0059] in, is the mass flow rate of chilled water in the dry cooling coil, c pw is the specific heat capacity of the chilled water in the dry cooling coil, T cw,in is the inlet temperature of the chilled water at the inlet of the dry cooling coil, T cw,out is the outlet temperature of the chilled water at the outlet of the dry cooling coil, The heat transferred by the chilled water in the dry cooling coil; is the air mass flow rate in the fresh air handling unit model, c pa is the specific heat capacity of air, T air,in Indicates the air temperature before passing through the dry cooling coil, T air,out Indicates the air temperature after passing through the dry cooling coil;

[0060] The transient energy balance equation of the heat transfer process corresponding to the dry cooling coil model is as follows:

[0061]

[0062] The transient energy balance equation of the wet cooling coil model of the fresh air handling unit model is as follows:

[0063]

[0064] h=c pa ·T air +ω·h fg ;

[0065] h in =c pa ·T in +ωin ·h fg ;

[0066] h out =c pa ·T out +ω out ·h fg ;

[0067] Among them, h in is the enthalpy of the air entering the wet cooling coil, h out is the enthalpy of the air leaving the wet cooling coil, h in -h out Contains the changes in sensible heat and latent heat in the wet cooling coil model, h is the enthalpy of air, T air is the air temperature, ω is the air humidity ratio, h fg is the latent heat of vaporization of water; T in is the air temperature entering the wet cooling coil, ω in is the humidity ratio of the air entering the wet cooling coil, T out is the air temperature leaving the wet cooling coil, ω out is the humidity ratio of the air leaving the wet cooling coil;

[0068] In the specific implementation process, a simulation model of the cleanroom HVAC system (including the air-side system and production workshop) was constructed based on the Modelica modeling language;

[0069] Step 2, performing a global sensitivity analysis on each input parameter of the constructed HVAC system model, respectively calculating the influence value of each input parameter on the cooling load output by the HVAC system model, and selecting multiple input parameters with an important influence degree on the cooling load output by the HVAC system model according to all the influence values ​​obtained, so as to form a sensitive input parameter set with the selected multiple input parameters; wherein the influence value of the input parameter on the cooling load output by the HVAC system model is positively correlated with its important influence degree;

[0070] Specifically in this embodiment, the sensitivity value of the cooling load output by the HVAC system model to any input parameter is expressed as follows:

[0071]

[0072] θ=(θ1,θ2,…,θ n );θ'=(θ1,θ2,…,θ i-1 ,θ i +△,θ i+1 ,…,θ n );

[0073]

[0074] Among them, d(θ i ) represents the cooling load output by the HVAC system model for the i-th input parameter θ i The sensitivity value, θ i is the i-th input parameter in the HVAC system model that affects the cooling load output by the HVAC system model, n is the total number of input parameters in the HVAC system model that affect the cooling load output by the HVAC system model, △ is the step size for perturbing the input parameters of the HVAC system model in the global sensitivity analysis, θ is the benchmark parameter set, θ' is the i-th input parameter θ in the benchmark parameter set θ i The new parameter set after being disturbed by a step length △; f(θ) represents the cooling load corresponding to the output when the input parameter set of the HVAC system model is the benchmark parameter set θ; f(θ') represents the cooling load corresponding to the output when the input parameter set of the HVAC system model is the new parameter set θ'; H represents the level of detail after each input parameter of the HVAC system model is discretized in the global sensitivity analysis;

[0075] Input parameter θ i The impact value on the cooling load output by the HVAC system model is marked as

[0076]

[0077] in, is the input parameter θ i The average effect value of the cooling load output by the HVAC system model, r is the total number of trajectories selected for global sensitivity analysis, d ij (θ) represents the jth trajectory among the r trajectories for the input parameter θ i The perturbation effect value of the basic effect, σ i is the input parameter θ i The standard deviation of the cooling load output by the HVAC system model;

[0078] For example, by presetting the impact value threshold Once the impact value of any input parameter on the cooling load output by the HVAC system model exceeds the impact value threshold The arbitrary input parameter is selected as an input parameter having a significant influence on the cooling load output by the HVAC system model;

[0079] It should be noted that sensitivity analysis helps to simplify the subsequent calibration process of the model and reduce the computational load during computer simulation;

[0080] Step 3: Based on the HVAC system model, the actual measurement data of cooling load in the preset historical period obtained in advance is used, and a PID controller is used to perform feedback adjustment and back-stepping on the missing input parameters in the sensitive input parameter set to obtain the missing input parameters obtained after back-stepping, and the missing input parameters obtained after back-stepping and all other input parameters in the sensitive input parameter set together form the training set data;

[0081] In this embodiment, the control formula of the PID controller used is as follows:

[0082]

[0083] Where IL(t) is the missing input parameter obtained after inference, e(t) is the error between the observed cooling load and the cooling load output by the constructed HVAC system model, and K p is the proportional gain, K i is the integral gain in the PID controller;

[0084] Step 4: Input the formed training set data and the sensitive input parameter set within any time period into the machine learning model for prediction processing, and predict the missing input parameters within the arbitrary time period; wherein the machine learning model is an XGBoos model, and the XGBoos model is used to perform machine learning inference on the missing input parameters in the sensitive input parameter set. The objective function of the XGBoos model is expressed as follows:

[0085]

[0086] Among them, L(φ) is the objective function of the XGBoost model, Represents the true value y of the vth sample v and the predicted value of the vth sample The loss function of the error between, K represents the total number of decision regression trees that form the XGBoost training model, and f k represents the kth decision regression tree among K decision regression trees, Ω(f k ) is the model complexity penalty term corresponding to the k-th decision regression tree;

[0087] For example, the input parameters of the machine learning model include outdoor temperature and solar radiation, and the inherent model parameters of the machine learning model are the U value of the wall and the heat exchange capacity of the air conditioning coil, etc.; the arbitrary time period selected here is assumed to be three days, and the training set data includes seven parameters: outdoor temperature, outdoor humidity, chilled water supply temperature, hot water supply temperature, medium-temperature water supply temperature, indoor temperature and outdoor temperature; after the prediction processing of the machine learning model, the missing input parameter in the arbitrary time period is predicted to be the internal heat gain data;

[0088] Step 5, using the remaining input parameters in the sensitive input parameter set and excluding the missing input parameter obtained by prediction to form a plurality of non-completely identical input parameter combinations, and inputting each input parameter combination into the HVAC system model respectively, and obtaining the HVAC system model output value corresponding to each input parameter combination respectively;

[0089] Assume that the sensitive input parameter set is labeled as α, α = {α1, α2, …, α n}, the predicted internal heat gain data of the missing input parameter is α in the sensitive input parameter set α n-1 , then, the "remaining input parameters" referred to here in step 5 are the sensitive input parameter set α excluding the sensitive input parameter α n-1 The remaining n-1 input parameters are obtained; then, different combinations are made for the n-1 input parameters to obtain different parameter combination forms; then, each obtained parameter combination is input into the HVAC system model as an input parameter set to obtain the HVAC system model output value corresponding to each input parameter combination;

[0090] Step 6, using a genetic algorithm tool to respectively calculate the output error value between each HVAC system model output value and the actual measurement data of the cooling load in the preset historical period, and select the input parameter combination corresponding to the output error value with the minimum error value as the optimal input parameter combination of the HVAC system model; wherein, using a genetic algorithm tool to calculate the output error between two values ​​is a conventional technical means in the art, and the specific process is not repeated here;

[0091] Step 7: input the actual value of the input parameter corresponding to each input parameter in the optimal input parameter combination into the HVAC system model, and use the cooling load output by the HVAC system model as the predicted value of the refrigeration load of the industrial clean room.

[0092] Although the preferred embodiments of the present invention are described in detail above, it should be clearly understood that various modifications and variations are possible for those skilled in the art. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the refrigeration load of an industrial clean room, characterized in that: The steps include: Step 1, constructing a HVAC system model including a room thermal model, a fresh air handling unit model, a dry cooling coil model, and a fan filter unit model; Step 2, performing a global sensitivity analysis on each input parameter of the constructed HVAC system model, respectively calculating the influence value of each input parameter on the cooling load output by the HVAC system model, and selecting multiple input parameters with an important influence degree on the cooling load output by the HVAC system model according to all the influence values ​​obtained, so as to form a sensitive input parameter set with the selected multiple input parameters; wherein the influence value of the input parameter on the cooling load output by the HVAC system model is positively correlated with its important influence degree; Step 3: Based on the HVAC system model, the actual measurement data of cooling load in the preset historical period obtained in advance is used, and a PID controller is used to perform feedback adjustment and back-stepping on the missing input parameters in the sensitive input parameter set to obtain the missing input parameters obtained after back-stepping, and the missing input parameters obtained after back-stepping and all other input parameters in the sensitive input parameter set together form the training set data; Step 4: Input the formed training set data and the sensitive input parameter set in any time period into the machine learning model for prediction processing, and predict the missing input parameters in the arbitrary time period; Step 5, using the remaining input parameters in the sensitive input parameter set and excluding the missing input parameter obtained by prediction to form a plurality of non-completely identical input parameter combinations, and inputting each input parameter combination into the HVAC system model respectively, and obtaining the HVAC system model output value corresponding to each input parameter combination respectively; Step 6, using a genetic algorithm tool to calculate the output error values ​​between the output values ​​of each HVAC system model and the actual measurement data of the cooling load in the preset historical period, and selecting the input parameter combination corresponding to the output error value with the minimum error value as the optimal input parameter combination of the HVAC system model; Step 7: input the actual value of the input parameter corresponding to each input parameter in the optimal input parameter combination into the HVAC system model, and use the cooling load output by the HVAC system model as the predicted value of the refrigeration load of the industrial clean room.

2. The method for predicting the refrigeration load of an industrial clean room according to claim 1, characterized in that: In step 1: The room thermal model uses a capacitor-resistance model to describe the dynamic change of indoor temperature. The transient energy balance equation of the room thermal model is as follows: Among them, C room is the heat capacity of the room, T room is the room temperature, T ext is the outdoor temperature of the room, R th is the thermal resistance of the room, Q int is the impact of the room’s internal heat load on the indoor temperature, represents the rate of change of the room's indoor temperature over time, Indicates the transient change value of the heat in the room. It indicates the influence of external heat conducted through thermal resistance on the indoor temperature of the room; The fresh air handling unit model includes a wet cooling coil model, a reheat coil model and a fan model. The fresh air handling unit model processes the sensible heat and latent heat in the air. The dry cooling coil model and the fan filter unit model together form a fan coil unit. The transient energy balance equation corresponding to the fan coil unit is as follows: in, is the mass flow rate of chilled water in the dry cooling coil, c pw is the specific heat capacity of the chilled water in the dry cooling coil, T cw,in is the inlet temperature of the chilled water at the inlet of the dry cooling coil, T cw,out is the outlet temperature of chilled water at the outlet of the dry cooling coil, The heat transferred by the chilled water in the dry cooling coil; is the air mass flow rate in the fresh air handling unit model, c pa is the specific heat capacity of air, T air,in Indicates the air temperature before passing through the dry cooling coil, T air,out Indicates the air temperature after passing through the dry cooling coil; The transient energy balance equation of the heat transfer process corresponding to the dry cooling coil model is as follows: The transient energy balance equation of the wet cooling coil model of the fresh air handling unit model is as follows: h=c pa ·T air +ω·h fg ; h in =c pa ·T in +ω in ·h fg ; h out =c pa ·T out +ω out ·h fg ; Among them, h in is the enthalpy of the air entering the wet cooling coil, h out is the enthalpy of the air leaving the wet cooling coil, h is the enthalpy of the air, T air is the air temperature, ω is the air humidity ratio, h fg is the latent heat of vaporization of water, T in is the air temperature entering the wet cooling coil, ω in is the humidity ratio of the air entering the wet cooling coil, T out is the air temperature leaving the wet cooling coil, ω out is the humidity ratio of the air leaving the wet cooling coil.

3. The method for predicting the refrigeration load of an industrial clean room according to claim 2, characterized in that: In step 2, the sensitivity of the cooling load output by the HVAC system model to any input parameter is expressed as follows: θ=(θ1,θ2,…,θ n );θ'=(θ1,θ2,…,θ i-1 ,i i +△,θ i+1 ,…,θ n ); Among them, d(θ i ) represents the cooling load output by the HVAC system model for the i-th input parameter θ i The sensitivity value, θ i is the i-th input parameter in the HVAC system model that affects the cooling load output by the HVAC system model, n is the total number of input parameters in the HVAC system model that affect the cooling load output by the HVAC system model, △ is the step size for perturbing the input parameters of the HVAC system model in the global sensitivity analysis, θ is the benchmark parameter set, θ' is the i-th input parameter θ in the benchmark parameter set θ i The new parameter set after being disturbed by a step length △; f(θ) represents the cooling load corresponding to the output when the input parameter set of the HVAC system model is the benchmark parameter set θ; f(θ') represents the cooling load corresponding to the output when the input parameter set of the HVAC system model is the new parameter set θ'; H represents the level of detail after each input parameter of the HVAC system model is discretized in the global sensitivity analysis; Input parameter θ i The impact value on the cooling load output by the HVAC system model is marked as in, is the input parameter θ i The average effect value of the cooling load output by the HVAC system model, r is the total number of trajectories selected for global sensitivity analysis, d ij (θ) represents the jth trajectory among the r trajectories for the input parameter θ i The perturbation effect value of the basic effect, σ i is the input parameter θ i The standard deviation of the effect on the cooling load output by the HVAC system model.

4. The method for predicting the refrigeration load of an industrial clean room according to claim 3, characterized in that: In step 3, the control formula of the PID controller is as follows: Where IL(t) is the missing input parameter obtained after inference, e(t) is the error between the observed cooling load and the cooling load output by the constructed HVAC system model, and K p is the proportional gain, K i is the integral gain in the PID controller.

5. The method for predicting the refrigeration load of an industrial clean room according to claim 4, characterized in that: In step 4, the machine learning model is an XGBoos model, and the XGBoos model is used to perform machine learning inference on the missing input parameters in the sensitive input parameter set. The objective function of the XGBoos model is expressed as follows: Among them, L(φ) is the objective function of the XGBoost model, Represents the true value y of the vth sample v and the predicted value of the vth sample The loss function of the error between, K represents the total number of decision regression trees that form the XGBoost training model, and f k represents the kth decision regression tree among K decision regression trees, Ω(f k ) is the model complexity penalty term corresponding to the kth decision regression tree.

Images