Predictive control method and device for fairing air-conditioning system

Abstract

The invention provides a predictive control method and device for a fairing air conditioning system. The predictive control method comprises the steps that actual state parameter values of the system are continuously collected; outputting prediction parameter values through a mechanism-data hybrid prediction model, a time sequence prediction model and a robust anti-interference prediction model, and determining a dynamic weight in a current control period; determining a fusion prediction parameter value of the current control period according to the prediction parameter value and the dynamic weight of each model; according to the fusion prediction parameter value of the current control period, a predefined multi-objective optimization function is solved, and the optimal control quantity of the current control period is obtained; the fairing air conditioning system is controlled according to the actual output value, the prediction error between the actual output value and the fusion prediction parameter value of the current control period is determined, and the fusion prediction parameter value of the next control period is corrected according to the prediction error; loop iteration is carried out in a new control period, rolling optimization control is carried out on the fairing air conditioning system, and the effects of being high in all-working-condition adaptability, excellent in anti-interference performance and high in control precision are achieved.

Description

Technical Field

[0001] This application relates to the field of precision environmental control technology, and in particular to a predictive control method and device for a fairing air conditioning system. Background Technology

[0002] In aerospace, precision electronics, and other fields, fairings serve as critical structural components protecting internal core equipment (such as satellite antennas and laser sensors). The temperature, humidity, and airflow velocity stability of their internal microenvironment directly affect the measurement accuracy, operational reliability, and lifespan of these devices. In practical applications, the enclosed internal space of the fairing, its small heat capacity, and low thermal inertia make it susceptible to disturbances from sudden changes in external solar radiation, drastic fluctuations in ambient temperature, and changes in the operating conditions of internal electronic equipment (such as startup, standby, and full load). This results in thermal and humidity loads exhibiting strong nonlinearity, time-varying characteristics, and multiple disturbance coupling features, posing a severe challenge to the high-precision environmental control of air conditioning systems.

[0003] Currently, rectifier-style air conditioning systems commonly employ traditional PID control or single-model predictive control (MPC) strategies. However, PID controllers rely on fixed parameter tuning, making it difficult to balance dynamic response speed and steady-state accuracy. They exhibit slow response to dynamic load changes and external disturbances, resulting in large overshoot, high steady-state error, and long settling time. Meanwhile, single-model predictive control struggles to accurately characterize the complex nonlinear dynamic characteristics of the system and has weak anti-interference capabilities. Especially under abrupt changes in operating conditions, model mismatch is severe, leading to a significant decline in control performance. Furthermore, while some existing research has proposed multimodal control, these methods often employ a hard-switching mechanism, switching between different control modes through statically preset rules. Modal transitions easily trigger control output shocks, causing system oscillations. Simultaneously, the lack of a collaborative fusion mechanism between modes fails to fully leverage complementary advantages and cannot adaptively adjust according to real-time operating conditions, resulting in limited overall control performance in complex dynamic environments. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a predictive control method and device for a fairing air conditioning system, which can solve the problems of poor adaptability, weak anti-disturbance capability and insufficient control accuracy of existing fairing air conditioning control algorithms, and achieve the effects of strong adaptability to all working conditions, excellent anti-disturbance performance and high control accuracy.

[0005] This application provides a predictive control method for a fairing air conditioning system, the method comprising: Continuously collect actual status parameter values ​​of the fairing air conditioning system; Based on the actual state parameter values, predicted parameter values ​​are output by each predicted model in a pre-built multimodal prediction model library; wherein, the multimodal prediction model library includes multiple prediction models; the prediction models include a mechanism-data hybrid prediction model, a time-series prediction model, and a robust disturbance rejection prediction model; the mechanism-data hybrid prediction model includes: establishing a heat and moisture transfer mechanism model in the fairing space based on the first law of thermodynamics and the principle of moisture balance; identifying unknown parameters in the heat and moisture transfer mechanism model in the fairing space using adaptive Kalman filtering; the time-series prediction model includes a deep learning model trained using historical time-series features; the robust disturbance rejection prediction model includes: based on... Control theory and disturbance observers are used to achieve robust control through state feedback and disturbance compensation. The dynamic weight of each prediction model in the current control cycle is determined based on the prediction parameter values ​​output by each prediction model. Based on the predicted parameter values ​​output by each prediction model and the dynamic weights in the current control cycle, determine the fusion predicted parameter values ​​for the current control cycle. Based on the fusion prediction parameter values ​​of the current control cycle, solve the predefined multi-objective optimization function to obtain the optimal control quantity for the current control cycle; The rectifier air conditioning system is controlled according to the optimal control quantity of the current control cycle, and the prediction error between the actual output value of the rectifier air conditioning system and the fusion prediction parameter value of the current control cycle is determined, so as to correct the fusion prediction parameter value of the next control cycle according to the prediction error. The system performs iterative cycles within the new control cycle to redetermine the optimal control quantity for the new control cycle, thereby achieving rolling optimization control of the rectifier air conditioning system.

[0006] This application embodiment also provides a predictive control device for a fairing air conditioning system, the device comprising: The data acquisition module is used to continuously collect the actual status parameter values ​​of the fairing air conditioning system; The prediction module is used to output predicted parameter values ​​based on the actual state parameter values ​​through each prediction model in a pre-built multimodal prediction model library. The multimodal prediction model library includes multiple prediction models, including a mechanism-data hybrid prediction model, a time-series prediction model, and a robust disturbance rejection prediction model. The mechanism-data hybrid prediction model includes: establishing a heat and moisture transfer mechanism model for the fairing space based on the first law of thermodynamics and the principle of moisture balance; and identifying unknown parameters in the heat and moisture transfer mechanism model for the fairing space using adaptive Kalman filtering. The time-series prediction model includes a deep learning model trained using historical time-series features. The robust disturbance rejection prediction model includes: based on... Control theory and disturbance observers are used to achieve robust control through state feedback and disturbance compensation. The first determining module is used to determine the dynamic weight of each prediction model in the current control cycle based on the prediction parameter values ​​output by each prediction model. The second determining module is used to determine the fusion prediction parameter value for the current control period based on the prediction parameter values ​​output by each prediction model and the dynamic weights in the current control period. The solution module is used to solve a predefined multi-objective optimization function based on the fusion prediction parameter values ​​of the current control cycle, and obtain the optimal control quantity for the current control cycle. The control module is used to control the rectifier air conditioning system according to the optimal control quantity of the current control cycle, determine the prediction error between the actual output value of the rectifier air conditioning system and the fusion prediction parameter value of the current control cycle, and correct the fusion prediction parameter value of the next control cycle according to the prediction error. The iterative module is used to perform cyclic iteration within a new control cycle to redetermine the optimal control quantity for the new control cycle, so as to achieve rolling optimization control of the rectifier air conditioning system.

[0007] This application provides a predictive control method and apparatus for a fairing air conditioning system. It constructs a multimodal predictive model library comprising a mechanism-data hybrid predictive model, a deep learning-based time-series predictive model, and a robust disturbance rejection predictive model. These models are specialized for steady-state, dynamic, and strong disturbance conditions, exhibiting strong adaptability to all operating conditions and robust anti-interference capabilities. Through a dynamic weight fusion mechanism in each control cycle, organic collaboration among multiple models is achieved, avoiding the control shock caused by traditional "hard switching" and improving control accuracy. Solving the multi-objective optimization function yields the optimal control quantity for the current control cycle, enabling collaborative optimization of multiple objectives and demonstrating excellent control performance.

[0008] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This application provides a flowchart of one of the predictive control methods for a fairing air conditioning system according to an embodiment of the present application. Figure 2 This paper illustrates the architecture of a predictive control system for a fairing air conditioning system provided in an embodiment of this application. Figure 3 This document shows a second flowchart of a predictive control method for a fairing air conditioning system provided in an embodiment of this application. Figure 4 This paper shows a schematic diagram of the structure of a predictive control device for a rectifier air conditioning system provided in an embodiment of this application; Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0012] Research has revealed that current air conditioning systems with fairings commonly employ traditional PID control or single-model predictive control (MPC) strategies. However, PID controllers rely on fixed parameter tuning, making it difficult to balance dynamic response speed and steady-state accuracy. They exhibit slow response to dynamic load changes and external disturbances, resulting in large overshoot, high steady-state error, and long settling time. Meanwhile, single-model predictive control uses only mechanistic or data-driven models. Mechanistic models struggle to accurately characterize the complex nonlinear dynamic characteristics of the system, while data-driven models have weak anti-interference capabilities, especially exhibiting severe model mismatch and significantly degraded control performance under abrupt changes in operating conditions. Furthermore, while some existing research has proposed multimodal control, these methods often employ a hard-switching mechanism, switching between different control modes through statically preset rules. Modal transitions easily trigger control output shocks, causing system oscillations. Simultaneously, the lack of a collaborative fusion mechanism between modes fails to fully leverage complementary advantages and cannot adaptively adjust according to real-time operating conditions, resulting in limited overall control performance in complex dynamic environments.

[0013] Based on this, the present application provides a predictive control method for a fairing air conditioning system to solve the problems of poor adaptability, weak anti-disturbance capability, and insufficient control accuracy of existing fairing air conditioning control algorithms, and achieve the effects of strong adaptability to all operating conditions, excellent anti-disturbance performance, and high control accuracy.

[0014] Please see Figure 1 , Figure 1 This is a flowchart illustrating a predictive control method for a fairing air conditioning system provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the method includes: S101. Continuously collect the actual status parameter values ​​of the fairing air conditioning system.

[0015] In this step, each control cycle can be... The actual state parameter values ​​of the fairing air conditioning system are collected once. In addition, the initial data can be preprocessed, such as by Kalman filtering, to suppress noise.

[0016] In practical implementation, the system can deploy multiple types of high-precision sensors as the sensing layer, such as temperature and humidity sensors and airflow velocity sensors inside the fairing; and ambient temperature sensors and solar radiation sensors outside. This allows for coverage of the microenvironmental parameters inside the fairing (temperature and humidity Tz / z), airflow velocity vs), external interference parameters (ambient temperature Tenv, solar radiation), and equipment operating parameters (load Qload) are used to ensure the comprehensiveness and accuracy of data collection.

[0017] S102. Based on the actual state parameter values, output the prediction parameter values ​​through each prediction model in the pre-built multimodal prediction model library.

[0018] The multimodal prediction model library includes multiple prediction models, including a mechanism-data hybrid prediction model, a time-series prediction model, and a robust disturbance-resistant prediction model. In other words, the multimodal prediction model library can include all three types of prediction models, or any two of them. Preferably, in this embodiment, the multimodal prediction model library includes all three types of prediction models, adapted to steady-state, dynamic, and strong disturbance conditions, respectively. The construction method of each prediction model will be described in detail below.

[0019] First, regarding the mechanism-data hybrid prediction model Based on the first law of thermodynamics and the principle of moisture balance, this application establishes a heat and moisture transfer mechanism model for the fairing space; and uses adaptive Kalman filtering to identify unknown parameters in the heat and moisture transfer mechanism model for the fairing space.

[0020] More specifically, the steps to build this model include: Step a1: Based on the first law of thermodynamics, the energy change of the air within the enclosed space of the fairing is equal to the difference between the input energy and the output energy, that is:

[0021] in, The total air energy within the enclosed space of the fairing ( For the heat capacity of air, (Temperature inside the fairing); Input energy Including the heat delivered by the air conditioner Equipment heat load Output energy This includes heat dissipated into the environment through the building envelope. Solar radiation heat dissipation Air convection heat transfer . Air supply - shroud internal temperature heat transfer coefficient Building envelope - ambient temperature heat transfer coefficient External ambient temperature, Air conditioning supply temperature.

[0022] The temperature dynamic equations in the fairing space heat and moisture transfer mechanism model are summarized as follows: .

[0023] Step a2: Based on the principle of moisture balance, the change in air humidity within the enclosed space of the fairing is equal to the difference between the moisture input and output, i.e.:

[0024] in, The total moisture content of the air inside the fairing ( Equivalent mass of air (Relative humidity inside the fairing); Input humidity. Including air conditioning supply humidity Equipment moisture dissipation Output humidity Including moisture permeability of the building envelope Moisture convection exchange rate ; The humidity transfer coefficient inside the air supply-rectifier hood; The relative humidity of the air supplied by the air conditioner.

[0025] The humidity dynamic equation in the above-mentioned fairing space heat and humidity transfer mechanism model is summarized as follows: .

[0026] Step a3: Define the unknown parameters that dynamically change with operating conditions in the fairing space heat and moisture transfer mechanism model as state variables, construct the state equation and observation equation of the adaptive Kalman filter, and identify the unknown parameters online through recursive formulas:

[0027] in, The unknown parameter is N; the number of samples is N; the convergence condition for parameter identification is: ; These are empirical parameter values; Let be the covariance matrix of the posterior estimate of the parameters at time k. For the kth The covariance matrix of the posterior estimate of the parameters at time 1.

[0028] Unknown parameters , , As the operating conditions change dynamically, they are treated as state variables, and the AKF state equations and observation equations are constructed as follows: Equations of state ,in, For AKF parameter vectors, Let k be the parameter vector. The noise of the AKF process at time k-1 follows a certain order. normal distribution, Process noise covariance matrix.

[0029] Observation equations ,in, Let k be the observation vector at time k. Given the input vector, h( ) Observation function, Let AKF be the observation noise at time k, which follows the order of... , Observation noise covariance matrix.

[0030] The AKF recursive formula includes a prediction step, covariance update, gain calculation, state update, and covariance correction. The prediction step is... Covariance updated to Gain calculation is Status updated to Covariance correction is .

[0031] The prior estimate (predicted value) of the parameter at time k is the initial inference of the parameter before combining it with the observation data at the current time. : kth The posterior estimate (optimal value) of the parameter at time 1 is the final accurate parameter obtained by combining the observation data from the previous time step. : Adaptive compensation term, the filter gain of AKF (calculated in advance through covariance), used to adjust the "correction strength of observation error on parameter prediction" (the larger the gain, the more sensitive the correction). : kth The actual observation value at time 1; : kth Theoretical observation at time 1; : The covariance matrix of the prior estimates of the parameters at time k (representing the uncertainty of the predicted parameter values, with the diagonal elements being the variances of each parameter); : kth The covariance matrix of the posterior estimate of the parameters at time 1 (corrected parameter uncertainty). Process noise covariance matrix (characterizing the statistical properties of internal system disturbances, such as equipment load fluctuations and random environmental disturbances). Kalman gain at time k (determines the degree of influence of measurement residuals on parameter correction, adaptively adjusted); : State of ; : Observation noise covariance matrix (characterizing measurement equipment errors, such as sensor accuracy and statistical characteristics of data acquisition noise); Finally, the expression for the mechanism-data hybrid prediction model is:

[0032] Where N is the number of samples, and the convergence condition for parameter identification is: These are empirical parameter values, for example. .

[0033] Second, regarding time series prediction models This includes deep learning models trained using historical time-series features; in this embodiment, a bidirectional gated recurrent unit (Bi-GRU) is used to alleviate the problem of long sequence dependencies through a gating mechanism, while capturing historical and future time-series features to improve prediction accuracy under dynamic conditions.

[0034] GRU resets the gate and Update Gate Control the flow of information.

[0035] Reset door Controlling the degree of forgetting historical information The function is a sigmoid function, with an output ∈ [0,1], where 0 represents complete forgetting. The weight matrix from the input layer to the reset gate. The weight matrix from the hidden layer to the reset gate. The hidden state vector of the GRU at time t-1, Reset the gate bias vector; Update Gate Control the ratio of historical information to current information. The weight matrix from the input layer to the update gate. The weight matrix from the hidden layer to the update gate. Update the gate bias vector; Candidate hidden state Based on the historical information after the reset gate filtering and the current input, calculations are performed. Hyperbolic tangent activation function, The weight matrix from the input layer to the candidate hidden state. The weight matrix from the hidden layer to the candidate hidden state. Candidate hidden state bias vector.

[0036] GRU forward hidden state vector at time t , The larger the value, the higher the weight of the current information.

[0037] Forward GRU captures historical time-series features Reverse GRU captures future time-series features By fusing bidirectional information through vector concatenation, the output can be predicted for the next H steps. ,in, For the GRU input vector, The weight matrix from the hidden layer to the output layer. The output layer bias vector is updated through training with the Adam optimizer, and the loss function uses mean squared error. ,in, The actual value of the i-th sample, The predicted value for the i-th sample.

[0038] Finally, the complete expression of the Bi-GRU model is:

[0039] Third, regarding robust disturbance rejection prediction models This includes: based on H∞ control theory and disturbance observers, treating external disturbances and load mutations as bounded disturbances, and achieving robust control through state feedback and disturbance compensation. The steps for constructing this model include: Step b1, System state-space modeling: Define the deviation state vector Control input Unknown interference vector , Using the system output vector, we perform state-space modeling to obtain the system state equations; where, Set the temperature value. Set the humidity value. Air conditioning airflow speed.

[0040] The system state equation is then:

[0041]

[0042] Wherein, the system state space matrix

[0043]

[0044]

[0045] C = I (identity matrix), E = 0.

[0046] Step b2 Derivation of robust performance constraints: Based on the system state equation Derivation of robust performance constraints requires unknown disturbance vectors To output transfer function satisfy Solve for the controller gain ;in, This refers to the anti-interference performance indicators.

[0047] but satisfy ,Right now, .

[0048] Solving for the controller gain using the linear matrix inequality (LMI) ,satisfy: in, It is a positive definite matrix.

[0049] in, The physical meaning is to traverse all interference frequencies. (e.g., instantaneous changes, slow fluctuations), calculate The maximum singular value, i.e. the maximum amplification factor of the interference at that frequency; Laplace operator for complex frequency variables , Determines the attenuation and amplification characteristics of the signal. The corresponding signal oscillation frequency.

[0050] Step b3, Derivation of the Disturbance Observer (DOB): Design an interference observer to estimate unknown interference. The observation equation is established as follows:

[0051]

[0052] in, For observer gain, Estimate the time derivative of the disturbance. Estimate the time derivative of the state, and estimate the state vector by designing the pole placement method (to minimize observation error). (Exponential convergence).

[0053] Ultimately, definition Given the disturbance compensation gain matrix, the robust disturbance rejection model control law is:

[0054] Step b4: Combine the system state equation, the observation equation, and the equation based on controller gain. and interference compensation gain matrix The determined robust disturbance rejection model control law is defined as the robust disturbance rejection prediction model. The complete model expression is: .

[0055] In step S102, the actual state parameter values ​​are input into each prediction model in the pre-built multimodal prediction model library to obtain the prediction parameter values ​​output by each prediction model.

[0056] S103. Determine the dynamic weight of each prediction model in the current control cycle based on the prediction parameter values ​​output by each prediction model.

[0057] In this embodiment, a dynamic weight calculation method based on real-time performance evaluation is designed, and the dynamic weight coefficients of the three prediction models are... and .

[0058] In one possible implementation, step S103 may include: S1031. For each prediction model, determine multiple performance evaluation index values ​​of the prediction model based on the prediction parameter values ​​output by the prediction model.

[0059] For example, multiple performance evaluation metrics include prediction accuracy, disturbance rejection performance, and output smoothness metrics.

[0060] Prediction accuracy metrics (root mean square error) , No. Predictive models (i.e., predictive models) The root mean square error of ). For predicting the time domain, for example ; For the first Step-by-step actual output value; For the first Prediction Model No. Step-by-step predicted value.

[0061] Disturbance immunity performance indicators Characterizes the interference suppression efficiency; Change in the output bias of the predictive model The amount of disturbance change .

[0062] Output smoothing index It is used to predict the degree of sequence fluctuation.

[0063] Next, to eliminate the influence of dimensions, the performance evaluation index values ​​can be normalized according to the index type (larger is better / smaller is better), as expressed by the formula:

[0064] in, No. Prediction Model No. Normalized index values ​​of each performance evaluation metric. , The first in historical data Maximum / minimum values ​​of the category indicator (obtained through offline statistics).

[0065] S1032. The combined weighting method is used to determine the combined weights corresponding to each evaluation index value of the prediction model.

[0066] In this step, firstly, the entropy weight method is used to determine the first... Entropy weights of performance evaluation metrics : The entropy weight method is used to calculate the first... Entropy value of each performance evaluation metric ;No. Prediction Model No. The percentage of normalized index values ​​for each performance evaluation indicator Finally, the entropy weights are calculated. .

[0067] Secondly, the CRITIC method is used to determine the objective weight of the j-th performance evaluation index of the prediction model. : The CRITIC method was used to calculate the first... Standard deviation of each performance evaluation indicator The average value of the normalized index of the j-th performance evaluation index Correlation coefficient between indicators ;No. Information content of each performance evaluation indicator ; final calculation of the first CRITIC objective weighting of each performance evaluation metric .

[0068] Finally, according to the first Entropy weights of performance evaluation metrics and objective weight The combined weights of the corresponding j-th evaluation index values ​​are determined as follows: ,satisfy .

[0069] S1033. Based on the multiple performance evaluation index values ​​of the prediction model and the combined weights corresponding to each evaluation index value, determine the dynamic weights of the prediction model in the current control cycle.

[0070] In this step, an exponential mapping is used to convert normalized index values ​​into weights, ensuring that models with better performance have larger weights.

[0071] In practical implementation, the dynamic weights of the prediction model in the current control cycle can be obtained by using the following formula, based on the multiple performance evaluation index values ​​of the prediction model and the combined weights corresponding to each evaluation index value, and employing the exponential mapping method:

[0072] in, Represents the prediction model Dynamic weights in the current control cycle; Represents the prediction model The normalized value of the j-th performance evaluation index.

[0073] S104. Determine the fusion prediction parameter value for the current control period based on the prediction parameter values ​​output by each prediction model and the dynamic weights in the current control period.

[0074] In this step, the fusion prediction parameter values ​​for the current control cycle can be determined by weighted summation. , ( ).

[0075] S105. Based on the fusion prediction parameter values ​​of the current control cycle, solve the predefined multi-objective optimization function to obtain the optimal control quantity for the current control cycle.

[0076] In one possible implementation, step S105 may include: S1051. Set the physical constraints of the fairing air conditioning system.

[0077] Considering the physical limitations of the air conditioning system and equipment safety thresholds, the constraints are as follows:

[0078] S1052. Introduce a barrier function to transform the optimization problem of the multi-objective optimization function under the physical constraints into an unconstrained problem.

[0079] The multi-objective optimization function is a weighted sum optimization function constructed by comprehensively considering four objectives: temperature and humidity accuracy, wind speed stability, energy consumption, and control smoothness. Specifically, it can be expressed as:

[0080] Temperature accuracy target Humidity accuracy target Wind speed stability target Energy consumption target ,in, Compressor energy consumption Fan energy consumption The energy consumption fitting coefficients a, b, c, d, and e were obtained through experimental fitting; the control smoothness target , ; , , , and These are the weights corresponding to the optimization control objectives, and the weights are... .

[0081] The constrained optimization problem is transformed into an unconstrained problem by introducing a barrier function:

[0082] in, This is the barrier parameter (which gradually decreases to 0 during the iteration process). For constraint functions (such as) ).

[0083] S1053. The obstacle function is solved iteratively using the interior point method. When the preset convergence condition is met, the optimal control quantity for the current control cycle is obtained.

[0084] Interior point method iterative steps: initialization Convergence accuracy .

[0085] gradient calculation .

[0086] Hessian matrix (second-order partial derivative matrix) calculation .

[0087] Newton steps solution .

[0088] Line search step size , .

[0089] The control input direction of the kth iteration , .

[0090] Convergence judgment, if Output the optimal control sequence Then the optimal control quantity for the current control cycle is .

[0091] S106. Control the rectifier air conditioning system according to the optimal control quantity of the current control cycle, determine the prediction error between the actual output value of the rectifier air conditioning system and the fusion prediction parameter value of the current control cycle, and correct the fusion prediction parameter value of the next control cycle according to the prediction error.

[0092] In this step, the PID control parameters are tuned; the parameter correction value is determined based on the prediction error and the PID control parameters; and the fusion prediction parameter value for the next control cycle is corrected based on the parameter correction value.

[0093] In practical implementation, prediction error , The actual output value is used to design a PID correction mechanism to adjust the prediction sequence and ensure prediction accuracy. The formula is expressed as:

[0094] The correction coefficients are tuned using the Ziegler-Nichols method, and the proportional coefficients are... ( (Critical gain); Integral coefficient ( (Critical period); differential coefficients .

[0095] S107. Perform iterative loops within the new control cycle to redetermine the optimal control quantity for the new control cycle, thereby achieving rolling optimization control of the rectifier air conditioning system.

[0096] Please see Figure 2 , Figure 2 This is an architectural diagram of a predictive control system for a fairing air conditioning system provided in an embodiment of this application. Figure 2 As shown. In this predictive control system.

[0097] Sensing layer: As the data input source, it deploys multiple types of high-precision sensors to cover microenvironmental parameters within the fairing (temperature, humidity, Tz / The data collection process includes parameters such as z (airflow velocity vs. airflow velocity), external interference parameters (ambient temperature Tenv, solar radiation), and equipment operating parameters (load Qload), ensuring comprehensiveness and accuracy. The sampling period is strictly controlled to 0.5s.

[0098] Data preprocessing module: Filters and reduces noise (Kalman filtering suppresses sensor noise) and removes outliers (removes data that is outside the measurement range or has abrupt changes) from the collected raw data, outputting smooth and reliable input vectors to provide high-quality data support for subsequent model predictions.

[0099] Multimodal prediction model library: Core model layer, three types of models with complementary functions and parallel computing: Mechanism-Data Hybrid Model ( Based on the first law of thermodynamics and the principle of moisture balance, the mechanism equation is derived, and the heat transfer / humidity coefficient is identified online through AKF to adapt to accurate prediction under steady-state conditions; Bi-GRU deep learning model ( ): Captures historical and future features of time-series data through a bidirectional gating unit to adapt to nonlinear prediction under dynamic operating conditions; Robust disturbance rejection model ( ): Combining H∞ control theory with a disturbance observer (DOB), it suppresses the effects of external disturbances and load mutations, and adapts to strong disturbance conditions.

[0100] Dynamic weight fusion module: Based on the real-time performance indicators (prediction accuracy, anti-disturbance performance, and output smoothness) of three types of models, the indicator weights are determined by the entropy weight-CRITIC combined weighting method, and then the dynamic weight coefficients ω1 / ω2 / ω3 are calculated by exponential mapping. The prediction results of the three types of models are weighted and fused to output a fused prediction sequence that balances accuracy and robustness.

[0101] Multi-objective rolling optimization module: Taking the fused prediction sequence as input, it constructs a multi-objective optimization function that includes temperature and humidity accuracy, wind speed stability, energy consumption, and control smoothness. Combining the system's physical constraints, it uses the interior-point method to solve for the optimal control sequence and outputs only the optimal control variables for the current period (supply air temperature Ts, humidity). s, wind speed vs).

[0102] PID feedback correction module: Calculates the deviation between the actual output and the fused prediction value, and corrects the prediction sequence of the next cycle through the PID correction mechanism (proportional Kp=0.6, integral Ki=0.1, derivative Kd=0.05) to make up for the model prediction deviation and improve the stability of the control closed loop.

[0103] The execution layer includes actuators such as variable frequency compressors, electronic expansion valves, variable speed fans, and ultrasonic humidifiers. It receives optimal control signals and executes them precisely to adjust the parameters of the rectifier air conditioning system.

[0104] Controlled object and feedback loop: The rectifier air conditioning system is the controlled object, and its output real-time microenvironment parameters (Tz / The z / vs signal is fed back to the perception layer, forming a complete control closed loop and ensuring continuous iterative optimization of the algorithm.

[0105] Please see Figure 3 , Figure 3 This is a second flowchart illustrating a predictive control method for a fairing air conditioning system provided in an embodiment of this application. Figure 3 As shown, the execution flow of this method includes: (1) Initialization: Load model parameters (AKF covariance) , Bi-GRU network weights; H∞ controller gain , ), initial weights Optimize weights and constraint thresholds, set prediction time domain H=20, and control period. ; (2) Data collection, each Data was collected once and preprocessed using a Kalman filter (to suppress noise). (3) Multimodal prediction: three types of models are computed in parallel, and the prediction sequence for the next 20 steps is output (computation time ≤ 0.3s). (4) Dynamic weight calculation: weights are updated based on real-time indicators (calculation time ≤ 0.05s); (5) Fusion prediction: The fusion sequence is obtained by weighted summation (time ≤ 0.02s); (6) Rolling optimization: Solve the optimization problem using the interior point method (time ≤ 0.1s), and execute the current control variable. ; (7) Feedback correction: Based on the actual output calculation error, correct the prediction sequence for the next period and return to step 2 loop.

[0106] In one embodiment of this application, the predictive control method for the fairing air conditioning system was used to conduct the following experiment: (a) Hardware system configuration 1. Perception layer The fairing is equipped with four sets of high-precision temperature and humidity sensors (measurement range: -20℃~60℃, accuracy ±0.1℃, ±1% RH) and two sets of airflow velocity sensors (measurement range: 0.1~5m / s, accuracy ±0.05m / s). Externally, it is equipped with an ambient temperature sensor and a solar radiation sensor. 2. Control Layer It adopts an STM32H743 microcontroller (480MHz main frequency, 1MB built-in SRAM), and integrates a 16-bit AD acquisition module, a PWM output module, and a CAN communication interface; 3. Execution layer Variable frequency compressor (cooling capacity 1-5kW), electronic expansion valve (opening adjustment range 0-100%), variable speed centrifugal fan (wind speed adjustment range 0.3-3m / s), ultrasonic humidifier (humidification capacity 0-5kg / h).

[0107] (II) Software Implementation 1. Development Environment MATLAB (model training and simulation), Keil MDK (embedded code development), Python (GRU model training).

[0108] 2. Model Training Collect 15,000 sets of operating data under different working conditions (steady-state, dynamic, and strong disturbance), with a training set / test set ratio of 8:2. The prediction error of the GRU model on the test set is ≤0.25℃ and ≤1.8% RH, and the convergence speed of the AKF identification parameters is ≤3s.

[0109] 3. Parameter settings Optimize target weights Correction coefficient , , Robust performance indicators

[0110] 4. Code porting The algorithm model was converted into C language code, ported to the STM32H743 microcontroller, and communicated with the actuator via the CAN bus to achieve real-time output of control signals.

[0111] (III) Testing and Verification 1. Steady-state operating condition test With an ambient temperature of 25℃, an equipment load of 3kW, a set temperature of 22℃, a humidity of 50% RH, and a wind speed of 0.8m / s, the steady-state error of temperature under algorithm control is ±0.12℃, and the steady-state error of humidity is ±1.2% RH, which is better than traditional PID control (±0.5℃, ±4% RH).

[0112] 2. Dynamic interference test When the ambient temperature suddenly changes from 25℃ to 35℃ (interference amplitude of 10℃), the algorithm response time is 3.1s and the overshoot is 2.5%, which significantly improves the anti-interference capability compared with the single GRU model (response time 5.2s, overshoot 7.8%).

[0113] 3. Load mutation test The equipment load suddenly increased from 3kW to 6kW (doubled), with temperature and humidity fluctuations ≤0.6℃ and ≤3% RH, and a recovery time ≤3.8s. Energy consumption was reduced by 15.6% compared to traditional MPC. 4. Long-term operation test After 24 hours of continuous operation, the system ran stably without any control shocks, and its average energy consumption was 23.4% lower than that of PID control.

[0114] This application provides a predictive control method for a fairing air conditioning system, which solves the problems of poor adaptability, weak anti-disturbance capability, and insufficient control accuracy of existing fairing air conditioning control algorithms. It provides a predictive control algorithm that integrates the advantages of multiple models, dynamically adjusts weights, and considers multiple control objectives. More specifically, this method has the following beneficial effects.

[0115] 1. Strong adaptability to all working conditions, significantly improved control accuracy: A multimodal prediction model library was constructed, comprising a mechanism-data hybrid prediction model, a deep learning-based time-series prediction model, and a robust disturbance-resistant prediction model. These three modal models cover steady-state, dynamic, and strongly disturbed scenarios, demonstrating strong adaptability across all operating conditions. A dynamic weight fusion mechanism enables organic collaboration and complementary advantages among the models, avoiding the control shocks caused by traditional "hard switching," resulting in a 20%-30% improvement in control accuracy compared to a single model. 2. Excellent anti-interference performance: A robust disturbance rejection prediction model based on H∞ control theory and disturbance observer (DOB) is introduced, and combined with a dynamic weighting mechanism to enhance its dominant role under sudden disturbances. This effectively suppresses temperature and humidity fluctuations caused by sudden external environmental changes (such as a 10℃ step temperature rise) and doubled internal load (2 times Qload sudden change), so that the system has a recovery time ≤4s and an overshoot ≤3% under sudden environmental changes of 10℃ and sudden load changes of 2 times.

[0116] 3. Multi-objective collaborative optimization: The design integrates a weighted multi-objective optimization function that considers temperature and humidity accuracy, wind speed stability, energy consumption, and control smoothness. It uses the interior point method to solve for the constrained optimal control sequence, minimizing the energy consumption of the air conditioning system while ensuring control quality. This achieves coordinated optimization of temperature and humidity accuracy, wind speed stability, and energy consumption, reducing energy consumption by 12%-18% compared to traditional MPC.

[0117] 4. High engineering practicality: The algorithm has moderate computational complexity (≤106 FLOPs per cycle) and can run in real time on embedded platforms such as STM32H7. It does not rely on high-performance computing hardware or cloud support and is suitable for practical deployment in resource-constrained scenarios such as spaceborne and airborne environments, and has good industrialization prospects.

[0118] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a predictive control device for a fairing air conditioning system provided in an embodiment of this application. Figure 4 As shown, the predictive control device 400 includes: The data acquisition module 410 is used to continuously acquire the actual status parameter values ​​of the fairing air conditioning system; Prediction module 420 is used to output predicted parameter values ​​based on the actual state parameter values ​​through each prediction model in a pre-built multimodal prediction model library; wherein, the multimodal prediction model library includes multiple prediction models; the prediction models include a mechanism-data hybrid prediction model, a time-series prediction model, and a robust disturbance rejection prediction model; the mechanism-data hybrid prediction model includes: establishing a heat and moisture transfer mechanism model in the fairing space based on the first law of thermodynamics and the principle of moisture balance; and identifying unknown parameters in the heat and moisture transfer mechanism model in the fairing space using adaptive Kalman filtering; the time-series prediction model includes a deep learning model trained using historical time-series features; the robust disturbance rejection prediction model includes: based on... Control theory and disturbance observers are used to achieve robust control through state feedback and disturbance compensation. The first determining module 430 is used to determine the dynamic weight of each prediction model in the current control cycle based on the prediction parameter values ​​output by each prediction model. The second determining module 440 is used to determine the fusion prediction parameter value for the current control period based on the prediction parameter values ​​output by each prediction model and the dynamic weights in the current control period. The solver module 450 is used to solve a predefined multi-objective optimization function based on the fusion prediction parameter values ​​of the current control cycle to obtain the optimal control quantity for the current control cycle. The control module 460 is used to control the rectifier air conditioning system according to the optimal control quantity of the current control cycle, determine the prediction error between the actual output value of the rectifier air conditioning system and the fusion prediction parameter value of the current control cycle, and correct the fusion prediction parameter value of the next control cycle according to the prediction error. The iteration module 470 is used to perform cyclic iteration within a new control cycle to redetermine the optimal control quantity for the new control cycle, so as to achieve rolling optimization control of the rectifier air conditioning system.

[0119] Furthermore, the prediction control device 400 further includes: a construction module; the construction module is used to construct the mechanism-data hybrid prediction model; the construction module is specifically used for: Based on the first law of thermodynamics, the energy change of air within the enclosed space of the fairing is equal to the difference between the input and output energy. The temperature dynamic equation in the heat and moisture transfer mechanism model of the fairing space is derived as follows:

[0120] in, For the heat capacity of air, Temperature inside the fairing, The heat transfer coefficient is the temperature inside the air supply and rectifier shroud. The air conditioning supply temperature, The heat transfer coefficient between the building envelope and the ambient temperature. External ambient temperature, For the equipment heat load, This refers to the heat dissipation caused by solar radiation. Heat exchange is achieved through air convection. Based on the principle of moisture balance, the change in air humidity within the enclosed space of the fairing is defined as equal to the difference between the moisture input and output. The dynamic equation for humidity in the heat and moisture transfer mechanism model of the fairing space is derived as follows:

[0121] in, Equivalent mass of air The relative humidity inside the fairing. The humidity transfer coefficient inside the air supply-rectifier shroud. To adjust the relative humidity of the air supplied by the air conditioner. For the moisture dissipation of the equipment, This refers to the moisture permeability of the building envelope. This refers to the amount of moisture convection exchange. The unknown parameters that dynamically change with operating conditions in the fairing space heat and moisture transfer mechanism model are defined as state variables. The state equation and observation equation of the adaptive Kalman filter are constructed, and the unknown parameters are identified online through recursive formulas.

[0122] in, The unknown parameter is N; the number of samples is N; the convergence condition for parameter identification is: , These are empirical parameter values; Let be the covariance matrix of the posterior estimate of the parameters at time k. For the kth The covariance matrix of the posterior estimate of the parameters at time 1.

[0123] Furthermore, the building module is also used to build the robust disturbance rejection prediction model; specifically, the building module is used for: Define the deviation state vector Control input Unknown interference vector , Using the system output vector, we perform state-space modeling to obtain the system state equations; where, Set the temperature value. Set the humidity value. Air conditioner airflow speed; Based on the system state equation Derivation of robust performance constraints requires unknown disturbance vectors To output transfer function satisfy Solve for the controller gain ;in, For anti-interference performance indicators; Design an interference observer to estimate unknown interference. Establish the observation equation and design the interference compensation gain matrix. ; The system state equation, the observation equation, and the controller gain-based equation are used. and interference compensation gain matrix The determined robust disturbance rejection model control law is defined as the robust disturbance rejection prediction model.

[0124] Furthermore, when the first determining module 430 determines the dynamic weight of each prediction model in the current control cycle based on the prediction parameter values ​​output by each prediction model, the first determining module 430 is used to: For each prediction model, multiple performance evaluation index values ​​are determined based on the prediction parameter values ​​output by the prediction model. The combined weighting method is used to determine the combined weights corresponding to each evaluation index value of the prediction model. Based on the multiple performance evaluation index values ​​of the prediction model and the combined weights corresponding to each evaluation index value, the dynamic weights of the prediction model in the current control cycle are determined.

[0125] Furthermore, when the first determining module 430 determines the combined weights corresponding to each evaluation index value of the prediction model using the combined weighting method, the first determining module 430 is used to: The entropy weight method is used to determine the first digit of the prediction model. Entropy weights of performance evaluation metrics ; The CRITIC method is used to determine the first prediction model. Objective weights of each performance evaluation metric ; According to the Entropy weights of performance evaluation metrics and objective weight Determine the corresponding first The combined weights of the evaluation index values ​​are: .

[0126] Furthermore, when the first determining module 430 determines the dynamic weights of the prediction model in the current control cycle based on multiple performance evaluation index values ​​of the prediction model and the combined weights corresponding to each evaluation index value, the first determining module 430 is used to: The dynamic weights of the prediction model in the current control cycle are obtained by using the following formula, based on the multiple performance evaluation index values ​​of the prediction model and the combined weights corresponding to each evaluation index value, and employing an exponential mapping method:

[0127] in, Represents the prediction model Dynamic weights in the current control cycle; Represents the prediction model No. Normalized index values ​​of each performance evaluation metric.

[0128] Furthermore, when the solving module 450 is used to solve a predefined multi-objective optimization function based on the fusion prediction parameter values ​​of the current control cycle to obtain the optimal control quantity for the current control cycle, the solving module 450 is used to: Define the physical constraints of the fairing air conditioning system; By introducing a barrier function, the optimization problem of the multi-objective optimization function under the physical constraints is transformed into an unconstrained problem; The obstacle function is solved iteratively using the interior point method. When the preset convergence condition is met, the optimal control quantity for the current control cycle is obtained.

[0129] Furthermore, the multi-objective optimization function is expressed as:

[0130] Among them, temperature accuracy target Humidity accuracy target Wind speed stability target Energy consumption target ,in, Compressor energy consumption Fan energy consumption The energy consumption fitting coefficients a, b, c, d, and e were obtained through experimental fitting; the control smoothness target , ; , , , and These are the weights corresponding to the optimization control objectives.

[0131] Furthermore, when the control module 460 corrects the fusion prediction parameter value for the next control cycle according to the prediction error, the control module 460 is used to: Tuning the PID control parameters; Based on the prediction error and the PID control parameters, determine the parameter correction value; The fusion prediction parameter values ​​for the next control cycle are corrected based on the parameter correction values.

[0132] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5 As shown, the electronic device 500 includes a processor 510, a memory 520, and a bus 530.

[0133] The memory 520 stores machine-readable instructions that can be executed by the processor 510. When the electronic device 500 is running, the processor 510 and the memory 520 communicate via the bus 530. When the machine-readable instructions are executed by the processor 510, the steps of the predictive control method for the rectifier air conditioning system as described in the above method embodiment can be executed. For specific implementation details, please refer to the method embodiment, which will not be repeated here.

[0134] This application also provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it can execute the steps of the predictive control method for the rectifier air conditioning system as described in the above method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0135] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0136] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0137] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0138] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0139] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0140] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A predictive control method for a fairing air conditioning system, characterized in that, The method includes: Continuously collect actual status parameter values ​​of the fairing air conditioning system; Based on the actual state parameter values, predicted parameter values ​​are output by each predicted model in a pre-built multimodal prediction model library; wherein, the multimodal prediction model library includes multiple prediction models; the prediction models include a mechanism-data hybrid prediction model, a time-series prediction model, and a robust disturbance rejection prediction model; the mechanism-data hybrid prediction model includes: establishing a heat and moisture transfer mechanism model in the fairing space based on the first law of thermodynamics and the principle of moisture balance; identifying unknown parameters in the heat and moisture transfer mechanism model in the fairing space using adaptive Kalman filtering; the time-series prediction model includes a deep learning model trained using historical time-series features; the robust disturbance rejection prediction model includes: based on... Control theory and disturbance observers are used to achieve robust control through state feedback and disturbance compensation. The dynamic weight of each prediction model in the current control cycle is determined based on the prediction parameter values ​​output by each prediction model. Based on the predicted parameter values ​​output by each prediction model and the dynamic weights in the current control cycle, determine the fusion predicted parameter values ​​for the current control cycle. Based on the fusion prediction parameter values ​​of the current control cycle, solve the predefined multi-objective optimization function to obtain the optimal control quantity for the current control cycle; The rectifier air conditioning system is controlled according to the optimal control quantity of the current control cycle, and the prediction error between the actual output value of the rectifier air conditioning system and the fusion prediction parameter value of the current control cycle is determined, so as to correct the fusion prediction parameter value of the next control cycle according to the prediction error. The system performs iterative cycles within the new control cycle to redetermine the optimal control quantity for the new control cycle, thereby achieving rolling optimization control of the rectifier air conditioning system.

2. The method according to claim 1, characterized in that, The steps for constructing the mechanism-data hybrid prediction model include: Based on the first law of thermodynamics, the energy change of air within the enclosed space of the fairing is equal to the difference between the input and output energy. The temperature dynamic equation in the heat and moisture transfer mechanism model of the fairing space is derived as follows: in, For the heat capacity of air, Temperature inside the fairing. The heat transfer coefficient is the temperature inside the air supply and rectifier shroud. The air conditioning supply temperature, The heat transfer coefficient between the building envelope and the ambient temperature. External ambient temperature, For the equipment heat load, This refers to the heat dissipation caused by solar radiation. Heat exchange is achieved through air convection. Based on the principle of moisture balance, the change in air humidity within the enclosed space of the fairing is defined as equal to the difference between the moisture input and output. The dynamic equation for humidity in the heat and moisture transfer mechanism model of the fairing space is derived as follows: in, Equivalent mass of air The relative humidity inside the fairing. The humidity transfer coefficient inside the air supply-rectifier shroud. To adjust the relative humidity of the air supplied by the air conditioner. For the moisture dissipation of the equipment, This refers to the moisture permeability of the building envelope. This refers to the amount of moisture convection exchange. The unknown parameters that dynamically change with operating conditions in the fairing space heat and moisture transfer mechanism model are defined as state variables. The state equation and observation equation of the adaptive Kalman filter are constructed, and the unknown parameters are identified online through recursive formulas. in, The unknown parameter is N; the number of samples is N; the convergence condition for parameter identification is: , These are empirical parameter values; Let be the covariance matrix of the posterior estimate of the parameters at time k. For the kth The covariance matrix of the posterior estimate of the parameters at time 1.

3. The method according to claim 2, characterized in that, The steps for constructing the robust disturbance rejection prediction model include: Define the deviation state vector Control input Unknown interference vector , Using the system output vector, we perform state-space modeling to obtain the system state equations; where, Set the temperature value. Set the humidity value. Air conditioner airflow speed; Based on the system state equation Derivation of robust performance constraints requires unknown disturbance vectors To output transfer function satisfy Solve for the controller gain ;in, For anti-interference performance indicators; Design an interference observer to estimate unknown interference. Establish the observation equation and design the interference compensation gain matrix. ; The system state equation, the observation equation, and the controller gain-based equation are used. and interference compensation gain matrix The determined robust disturbance rejection model control law is defined as the robust disturbance rejection prediction model.

4. The method according to claim 1, characterized in that, The dynamic weights of each prediction model in the current control cycle are determined based on the prediction parameter values ​​output by each prediction model, including: For each prediction model, multiple performance evaluation index values ​​are determined based on the prediction parameter values ​​output by the prediction model. The combined weighting method is used to determine the combined weights corresponding to each evaluation index value of the prediction model. Based on the multiple performance evaluation index values ​​of the prediction model and the combined weights corresponding to each evaluation index value, the dynamic weights of the prediction model in the current control cycle are determined.

5. The method according to claim 4, characterized in that, The combined weighting method is used to determine the combined weights corresponding to each evaluation index value of the prediction model, including: The entropy weight method is used to determine the first digit of the prediction model. Entropy weights of performance evaluation metrics ; The CRITIC method is used to determine the first prediction model. Objective weights of each performance evaluation metric ; According to the Entropy weights of performance evaluation metrics and objective weight Determine the corresponding first The combined weights of the evaluation index values ​​are: .

6. The method according to claim 5, characterized in that, Based on the multiple performance evaluation index values ​​of the prediction model and the combined weights corresponding to each evaluation index value, the dynamic weights of the prediction model in the current control cycle are determined, including: The dynamic weights of the prediction model in the current control cycle are obtained by using the following formula, based on the multiple performance evaluation index values ​​of the prediction model and the combined weights corresponding to each evaluation index value, and employing an exponential mapping method: in, Represents the prediction model Dynamic weights in the current control cycle; Represents the prediction model No. Normalized index values ​​of each performance evaluation metric.

7. The method according to claim 1, characterized in that, Based on the fusion prediction parameter values ​​of the current control cycle, a predefined multi-objective optimization function is solved to obtain the optimal control quantity for the current control cycle, including: Define the physical constraints of the fairing air conditioning system; By introducing a barrier function, the optimization problem of the multi-objective optimization function under the physical constraints is transformed into an unconstrained problem; The obstacle function is solved iteratively using the interior point method. When the preset convergence condition is met, the optimal control quantity for the current control cycle is obtained.

8. The method according to claim 7, characterized in that, The multi-objective optimization function is expressed as: Among them, temperature accuracy target Humidity accuracy target Wind speed stability target Energy consumption target ,in, Compressor energy consumption Fan energy consumption The energy consumption fitting coefficients a, b, c, d, and e were obtained through experimental fitting; the control smoothness target , ; , , , and These are the weights corresponding to the optimization control objectives.

9. The method according to claim 1, characterized in that, The fusion prediction parameter values ​​for the next control cycle are corrected according to the prediction error, including: Tuning the PID control parameters; Based on the prediction error and the PID control parameters, determine the parameter correction value; The fusion prediction parameter values ​​for the next control cycle are corrected based on the parameter correction values.

10. A predictive control device for a fairing air conditioning system, characterized in that, The device includes: The data acquisition module is used to continuously collect the actual status parameter values ​​of the fairing air conditioning system; The prediction module is used to output predicted parameter values ​​based on the actual state parameter values ​​through each prediction model in a pre-built multimodal prediction model library. The multimodal prediction model library includes multiple prediction models, including a mechanism-data hybrid prediction model, a time-series prediction model, and a robust disturbance rejection prediction model. The mechanism-data hybrid prediction model includes: establishing a heat and moisture transfer mechanism model for the fairing space based on the first law of thermodynamics and the principle of moisture balance; and identifying unknown parameters in the heat and moisture transfer mechanism model for the fairing space using adaptive Kalman filtering. The time-series prediction model includes a deep learning model trained using historical time-series features. The robust disturbance rejection prediction model includes: based on... Control theory and disturbance observers are used to achieve robust control through state feedback and disturbance compensation. The first determining module is used to determine the dynamic weight of each prediction model in the current control cycle based on the prediction parameter values ​​output by each prediction model. The second determining module is used to determine the fusion prediction parameter value for the current control period based on the prediction parameter values ​​output by each prediction model and the dynamic weights in the current control period. The solution module is used to solve a predefined multi-objective optimization function based on the fusion prediction parameter values ​​of the current control cycle, and obtain the optimal control quantity for the current control cycle. The control module is used to control the rectifier air conditioning system according to the optimal control quantity of the current control cycle, determine the prediction error between the actual output value of the rectifier air conditioning system and the fusion prediction parameter value of the current control cycle, and correct the fusion prediction parameter value of the next control cycle according to the prediction error. The iterative module is used to perform cyclic iteration within a new control cycle to redetermine the optimal control quantity for the new control cycle, so as to achieve rolling optimization control of the rectifier air conditioning system.

Images