Clean air conditioning control method based on neural network and MPC algorithm

Abstract

The present invention relates to the field of neural network technology, and proposes a clean air conditioning control method based on a neural network and an MPC algorithm, including: collecting multiple dimensions of indoor and outdoor environmental data, obtaining multiple dimensions of state data and operation data of the air conditioning system; constructing an LSTM neural network model, based on the predicted values ​​of environmental parameters and environmental parameters at adjacent moments, and obtaining the power change trend of the air conditioning system by training the LSTM neural network model; setting the target environmental parameter range of each dimension based on the indoor clean conditions, constructing the energy consumption cost function of each dimension in the future time, and then optimizing and solving to obtain the optimal control input sequence of each dimension; controlling the air conditioning system based on the optimal control input sequence of each dimension. The present invention aims to solve the problem of energy waste and fixed adjustment caused by traditional air conditioning in a clean environment through fixed parameter control.

Description

Technical Field

[0001] The present invention relates to the technical field of neural network, and in particular to a clean air conditioning control method based on neural network and MPC algorithm. Background Art

[0002] Soft clean energy-saving air-conditioning technology is one of the key issues in the current field of building environment control. In order to improve the efficiency of air-conditioning systems and reduce energy consumption, researchers are committed to finding new and smarter control strategies. Traditional air-conditioning system control models are often difficult to adapt to dynamically changing environmental conditions. Therefore, an advanced prediction model that can flexibly respond to different working conditions and efficiently save energy is needed.

[0003] The energy-saving control of traditional air-conditioning systems in clean environments often relies on empirical rules or simple PID controllers, which lack flexibility and intelligent adaptability to dynamic changes in the environment. Therefore, an advanced prediction model that can combine neural networks and MPC algorithms is needed to more accurately predict future system states, achieve intelligent control, and minimize energy waste. Summary of the invention

[0004] The present invention provides a clean air conditioning control method based on a neural network and an MPC algorithm to solve the problem of energy waste and fixed adjustment caused by fixed parameter control of existing traditional air conditioners in a clean environment. The technical solutions adopted are as follows:

[0005] The present invention proposes a clean air conditioning control method based on a neural network and an MPC algorithm, the method comprising the following steps:

[0006] Collect multi-dimensional environmental data indoors and outdoors, and obtain multi-dimensional status data and operation data of the air-conditioning system;

[0007] Construct an LSTM neural network model, input the environmental parameters of each dimension at each moment, and output the predicted values ​​of the environmental parameters of each dimension at the next moment. Based on the predicted values ​​of the environmental parameters and the environmental parameters at adjacent moments, the power change trend of the air-conditioning system is obtained by training the LSTM neural network model.

[0008] The target environmental parameter range of each dimension is set based on the indoor clean conditions. Combined with the control input sequence and power change trend of each dimension, and the predicted value of each dimension of environmental parameters output by the LSTM neural network model, the energy consumption cost function of each dimension in the future time is constructed, and then the optimal control input sequence of each dimension is obtained by optimization.

[0009] The air conditioning system is controlled based on the optimal control input sequence in each dimension.

[0010] Preferably, the specific method of constructing the LSTM neural network model includes:

[0011] definition Layer LSTM neural network model, each layer has M neural units;

[0012] Initialize network weights and paranoia vector ;

[0013] Construct a loss function based on model parameters according to the input and output of the LSTM neural network model;

[0014] Use the back-propagation algorithm to calculate the gradient of the loss function with respect to each weight;

[0015] Apply gradient descent method to update model parameters;

[0016] Train an iterative LSTM neural network model.

[0017] Preferably, the specific method of obtaining the predicted values ​​of the environmental parameters of each dimension at the next moment includes:

[0018] According to Dataset at time Constructing the input feature vector , the dataset Including Environmental parameters of each dimension at each moment;

[0019] Output vector , based on the output vector Prediction dataset for each moment , the prediction data set Including The predicted values ​​of environmental parameters in each dimension at each moment.

[0020] Preferably, the loss function based on the model parameters is constructed according to the input and output of the LSTM neural network model, including the specific method of:

[0021] ;

[0022] in, Represents the number of samples in the training set; represents the model parameters; Indicates The input feature vector at time instant, Indicates The output vector at each moment; represents the loss function.

[0023] Preferably, the specific method of obtaining the power change trend of the air conditioning system includes:

[0024] ;

[0025] in, Indicates The input feature vector at time instant, Indicates The prediction data set at each moment, Indicates The output vector at each moment is Indicates Control input at each moment;

[0026] The control inputs at all times constitute the control input sequence , and then get the power change trend of the corresponding air conditioning system .

[0027] Preferably, the target environmental parameter ranges of various dimensions are set based on the indoor cleanliness conditions, and the specific method includes:

[0028] For any dimension of environmental parameters, set the target environmental parameter range .

[0029] Preferably, the construction of the energy consumption cost function in each dimension in the future time includes the following specific methods:

[0030] Set the future time and obtain the predicted data set for each moment in the future based on the LSTM neural network model; based on the power change trend of the air-conditioning system and the predicted data set for each moment in the future, construct the energy consumption cost function as follows:

[0031] ;

[0032] in, represents the energy consumption cost function of air conditioning; Indicated in At each moment, using the power change trend function , input control sequence Energy consumption of air conditioning; Indicates Predicted indoor environmental parameters at each moment; A reference parameter representing the environmental parameter of this dimension; and represents the weight factor; is the control input sequence of the air conditioner.

[0033] Preferably, the method of setting the future time and obtaining the prediction data set for each moment in the future time based on the LSTM neural network model includes the following specific methods:

[0034] Get the future through LSTM neural network model A dataset of predictions at each moment in time.

[0035] Preferably, the weight factor is obtained by:

[0036] Energy consumption weight ; Environmental parameter weight ;in, is the weight adjustment parameter, is the weight adjustment factor;

[0037] Calculate energy consumption error and environmental parameter error at each moment in the future;

[0038] The weight adjustment factor is adjusted according to the energy consumption error and environmental parameter error. The specific calculation process is:

[0039] ;

[0040] in, represents the weight adjustment factor, represents the number of future moments, and Respectively represent Energy consumption error and environmental parameter error at each moment, Represents the sigmoid function.

[0041] Preferably, the specific method of obtaining the optimal control input sequence of each dimension includes:

[0042] Energy consumption error ,in and Respectively represent The predicted energy consumption and actual energy consumption of the LSTM model of the air conditioning system at each moment;

[0043] Environmental parameter error ,in and Respectively represent The predicted value of the environmental parameters and the actual value of the environmental parameters of the LSTM model at each moment.

[0044] The beneficial effects of the present invention are as follows: the present invention predicts environmental parameters of each dimension through an LSTM neural network model, constructs a power change trend of each dimension, and then analyzes the energy consumption of the air-conditioning system in the future time, and optimizes and solves the difference between the predicted value of the environmental parameter in the future time and the requirement of the clean environment to obtain the optimal control input sequence of each dimension, so as to control the air-conditioning system; wherein, by constructing an LSTM neural network model, and inputting the environmental parameters of each dimension at the previous moment into the model as a data set, the predicted value and the true value of the next moment are analyzed based on the output, and the neural network model is trained, so as to obtain the control input of the environmental parameters specific to each dimension, and then obtain the power change trend of the air-conditioning system, and provide a basis for the subsequent air-conditioning system to achieve the control of the clean environment; the optimal control input sequence of the air-conditioning system under the environmental parameters specific to each dimension within a certain period of time in the future is obtained, so that the environmental parameters of each dimension in the room meet the target environmental parameter range, that is, meet the requirements of the clean environment, and ensure the minimum energy consumption; finally, the air-conditioning system adjusts the clean environment while the air-conditioning system consumes the minimum energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0046] Figure 1 A schematic flow chart of a clean air conditioning control method based on a neural network and an MPC algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] The purpose of the present invention is to analyze indoor and outdoor environmental data, use neural networks to predict future indoor and outdoor environments, and energy consumption trends corresponding to different control input sequences of air conditioners in the current environment; and then use the prediction results to solve the minimum energy consumption control sequence of fixed cleanliness standards under different environmental parameters.

[0049] The specific scenario targeted by the present invention is: in places with high requirements for cleanliness, such as laboratories, medical facilities, and production workshops, air conditioners need to be used to adjust the cleanliness of the places, and the existing adjustment method is to use environmental data obtained by sensors to implement adjustments. This adjustment method has no obvious sensitivity to environmental changes, resulting in the lack of flexibility and intelligent adaptability of air conditioning control to dynamic changes in the environment when the environment changes. Therefore, the present invention proposes to combine the LSTM neural network prediction model, and on the basis of the LSTM neural network prediction model, solve the minimum energy consumption control sequence of the MPC algorithm applied to the fixed cleanliness standard under different environmental parameters of the air conditioner. By utilizing the complex nonlinear characteristics of the LSTM neural network learning system and combining it with the MPC algorithm, accurate prediction of the future state of the system and optimal control of the minimum energy consumption can be achieved.

[0050] See also Figure 1 , which shows a flow chart of a clean air conditioning control method based on a neural network and an MPC algorithm provided by an embodiment of the present invention, the method comprising the following steps:

[0051] Step S001: Collecting multiple dimensions of indoor and outdoor environmental data, and obtaining multiple dimensions of status data and operation data of the air-conditioning system.

[0052] It should be noted that when air conditioners are used in laboratories, medical facilities, production workshops and other places with high requirements for cleanliness, it is necessary to use the indoor and outdoor environment and the system operating parameters of the air conditioner to analyze the cleanliness of the current working location, and then adjust the air conditioner parameters through intelligent control algorithms for subsequent intelligent control. Therefore, before this, it is necessary to first collect and pre-process the indoor and outdoor environmental data, and at the same time collect and pre-process the status data and operation data of the air conditioning system.

[0053] Indoor and outdoor environment data collection:

[0054] Collecting multiple dimensions of indoor and outdoor environmental data through indoor and outdoor sensor networks:

[0055] 1) Indoor: temperature (Tin), humidity (Hin), CO2 concentration (Cin);

[0056] 2) Outdoor: temperature (Tout), humidity (Hout), wind speed (Wout).

[0057] Air conditioning system status data and operation data collection:

[0058] Collect status and operation data of the air conditioning system through the built-in sensors and controller interface of the air conditioning system:

[0059] 1) Status data: on / off status (Sac, expressed as 0 or 1), set temperature (Tset), wind speed (Fac);

[0060] 2) Operation data: cold and hot source temperature (Tsource), cold and hot source flow (Fsource), fan current (Ifan), fan voltage (Vfan), fan speed (RPMfan), fan operating frequency (Ffan).

[0061] Data collection frequency:

[0062] 1) Indoor and outdoor environmental data: preset collection time interval , this embodiment sets the range , without any specific limitation;

[0063] 2) Status data and operation data of air conditioning system: preset collection time interval , this embodiment sets the range , without specific limitation.

[0064] It should be noted that the data collection process in this embodiment starts when the air-conditioning system starts running, and the data is collected in real time; then the collected data is cleaned and preprocessed. Taking temperature data as an example, the rest of the data cleaning method is the same as the temperature data cleaning. The data preprocessing process is to uniformly process the data of the same environmental parameters or the same dimension.

[0065] Data cleaning:

[0066] 1) First remove or interpolate missing values; if a data point is missing, use the average of the previous and next data points for interpolation;

[0067] 2) Identify and process outliers; if a data point exceeds a preset threshold range, mark the data point as an outlier and process it by deleting the data point and interpolating it using the adjacent data points before and after; the threshold ranges of environmental data, status data, and operating data of each dimension can be set using existing methods, and outliers can be detected, which will not be repeated in this embodiment.

[0068] Data preprocessing:

[0069] 1) Standardization to ensure that the data has zero mean and unit variance;

[0070] 2) Normalization, scaling the data to within the interval;

[0071] 3) Use a sliding average filter to smooth the data.

[0072] At this point, data collection and processing of environmental data, status data and operation data in multiple dimensions have been completed, and the processed environmental data are recorded as environmental parameters of each dimension, and there are environmental parameters at multiple times; the processed status data and operation data of each dimension are recorded as status parameters and operation parameters of each dimension, and there are also status parameters and operation parameters at multiple times.

[0073] Step S002, construct an LSTM neural network model, input the environmental parameters of each dimension at each moment, output the predicted values ​​of the environmental parameters of each dimension at the next moment, and obtain the power change trend of the air-conditioning system by training the LSTM neural network model based on the predicted values ​​of the environmental parameters and the environmental parameters at adjacent moments.

[0074] It should be noted that when the air conditioner adjusts parameters to achieve the purpose of indoor cleanliness, the implementation of the adjustment is likely to cause the indoor environmental adjustment to change rapidly, and because of the changeable characteristics of the indoor environment (for example, factories and laboratories process objects, which makes the indoor environment change greatly), the air conditioner has a certain lag when adjusting parameters in real time, and there is a phenomenon of frequent adjustment and energy consumption. Therefore, the LSTM neural network model is used as a prediction algorithm to predict indoor and outdoor environmental parameters, and based on the predicted environmental parameters, the change trend of the air conditioning system power in the future is predicted if the environmental parameters required for the clean area are to be met.

[0075] 1) Definition Layer LSTM neural network model, each layer has M neural units:

[0076] a) Input layer: According to Dataset at time Constructing the input feature vector , dataset Including Environmental parameters of each dimension at each moment;

[0077] b) Output layer: output vector , based on the output vector Prediction dataset for each moment , including The predicted values ​​of environmental parameters in each dimension at each moment.

[0078] 2) Model initialization: Use the Glorot initialization method to initialize network weights and paranoia vector .

[0079] 3) Forward propagation:

[0080] In the At this moment, for the input feature vector The following operations are performed through the network:

[0081] ;

[0082] in, Indicates The input feature vector at time instant, Indicates The prediction data set at each moment, Indicates The output vector at each moment is Indicates Control input at each moment;

[0083] 4) Training hyperparameters:

[0084] a) Use mean square error (MSE) as the loss function to evaluate the predicted value and the true value (No. The difference between the input feature vectors at time moments:

[0085] ;

[0086] in, Represents the number of samples in the training set. , Indicates the number of all collected moments. Taking the indoor temperature environmental parameter as an example, The temperature data is used as the training set for training, and the remaining The temperature data is used as a validation set for verification; represents the model parameters; Indicates The input feature vector at time instant, Indicates The output vector at each moment; represents the loss function.

[0087] b) Back propagation and parameter update:

[0088] Use the back-propagation algorithm to calculate the gradient of the loss function with respect to each weight;

[0089] Apply gradient descent to update model parameters:

[0090] ;

[0091] in, represents the model parameters before updating, represents the updated model parameters, It means that the loss function takes partial derivatives of the model parameters. Represents the learning rate.

[0092] c) Training hyperparameters:

[0093] In this embodiment, the learning rate The initial value of is set to 0.001;

[0094] The number of training iterations (epochs) is determined based on the performance of the LSTM neural network model on the validation set, using existing technology, and will not be described in detail in this embodiment.

[0095] 5) Model evaluation and validation:

[0096] a) After each training iteration, the performance of the LSTM neural network model is evaluated using a validation set, which is a prior art and will not be described in detail in this embodiment;

[0097] b) Monitor the validation loss and stop training if you notice overfitting (i.e. validation loss starts to increase).

[0098] Furthermore, through the constructed LSTM neural network model, the control input at each moment under any specific environmental parameters is obtained, and the control input at all moments constitutes the control input sequence , the specific environmental parameters are the environmental parameters required for a clean environment, and the power change trend of the corresponding air-conditioning system is obtained .

[0099] At this point, by constructing an LSTM neural network model and inputting the environmental parameters of each dimension at the previous moment into the model as a data set, the predicted value and the actual value at the next moment are analyzed based on the output, and the neural network model is trained to obtain the control input of the specific environmental parameters of each dimension, and then the power change trend of the air-conditioning system is obtained, which provides a basis for the subsequent air-conditioning system to achieve clean environment control.

[0100] Step S003: Set the target environmental parameter range of each dimension based on the indoor cleanliness conditions, combine the control input sequence and power change trend of each dimension, and the predicted value of the environmental parameter of each dimension output by the LSTM neural network model, construct the energy consumption cost function of each dimension in the future, and then optimize and solve to obtain the optimal control input sequence of each dimension.

[0101] It should be noted that the change trend obtained by prediction is a multivariate adjustable function, not the control effect obtained at the minimum cost under the most ideal state. Therefore, taking power consumption as the starting point, the formalized MPC optimization problem is used to solve the above Perform optimization to ensure prediction of future time The cost minimization problem within .

[0102] It should be further explained that, first, a control target is defined. The control target of this embodiment is to minimize the energy consumption range while maintaining the indoor environmental parameters within the clean area target range.

[0103] 1) Taking any dimension of environmental parameters as an example, set the target environmental parameter range ;

[0104] 2) Obtain the future through the LSTM neural network model A dataset of predictions at each moment in time, i.e. the predicted system behavior;

[0105] 3) Set the control horizon , is the time range of future control behavior, that is, the air conditioning cleanliness under the minimum energy consumption obtained by using the MPC model is Optimization control is performed within a certain time period, , Represents the number of all moments collected;

[0106] 4) Setting the objective function. Since the goal of this embodiment is to minimize energy consumption while ensuring that indoor environmental parameters meet the standards of the clean area, the objective function is set to the energy consumption cost. The objective function, i.e., the energy consumption cost function, is as follows:

[0107] ;

[0108] in, represents the energy consumption cost function of air conditioning; Indicated in At each moment, using the power change trend function , input control sequence Energy consumption of air conditioning; Indicates Predicted indoor environmental parameters at each moment; A reference parameter representing the environmental parameter of this dimension; and represents the weight factor; is the control input sequence of the air conditioner.

[0109] It should be noted that the difference between the predicted environmental parameter values ​​output by the LSTM neural network model and the reference parameters is as small as possible under the target environmental parameter range of this dimension. At the same time, the air conditioning energy consumption obtained by adjusting the control input at the corresponding moment through the power change trend function is reduced to minimize the air conditioning energy consumption, thereby completing the optimization solution.

[0110] 5) Set constraints. In this embodiment, the constraints consist of two parts. The first part is that the indoor environment must be within the indoor target environment parameter range. The other condition is that the control input at the future moment must satisfy the control input sequence. The constraints are set as:

[0111] a) ;

[0112] b) ;in and The control input sequence is The minimum and maximum values ​​in .

[0113] 6) Solve the optimization problem. Based on the set constraints, optimize the objective function. The solution method uses the numerical optimal algorithm SQP to solve the objective function and obtain the optimal solution at each moment in the future. The optimal control input sequence is obtained .

[0114] 7) Control and update the air conditioning system; obtain the The optimal control input at the moment, real-time data is updated to the At each moment, the predicted values ​​of the environmental parameters are updated based on the LSTM neural network model, and the objective function is updated to solve the optimization problem of the updated objective function, and so on.

[0115] 8) Set the parameters in the objective function;

[0116] a) Energy consumption weight ;

[0117] b) Environmental parameter weights ;

[0118] c) In this embodiment, the weight adjustment parameter preset value ; Preset value of weight adjustment factor .

[0119] 9) Error analysis;

[0120] a) First, in each , calculate the prediction error:

[0121] Energy consumption error ,in and Respectively represent The predicted energy consumption and actual energy consumption of the LSTM model of the air conditioning system at each moment;

[0122] Environmental parameter error ,in and Respectively represent The predicted value of environmental parameters and the actual value of environmental parameters of the LSTM model at each moment;

[0123] It should be noted that the difference between the predicted value and the actual value represents the error of the system.

[0124] b) Adjustment of control strategy:

[0125] Using prediction error to adjust weight adjustment factor , adjust the energy consumption weight and environmental parameter weight, the specific calculation process is:

[0126] ;

[0127] in, represents the weight adjustment factor, represents the number of future moments, and Respectively represent Energy consumption error and environmental parameter error at each moment, represents a sigmoid function, which is used for normalization processing in this embodiment.

[0128] It should be noted that the greater the cumulative error of energy consumption weight is compared with the cumulative error of environmental parameter weight, the greater the The larger it is, the more emphasis should be placed on the calculation of environmental parameters when calculating the optimization function, and vice versa.

[0129] Furthermore, the indoor temperature in the future is obtained by the above method. time, the optimal control input sequence of specific environmental parameters, and obtain the optimal control input sequence for each dimension of the room according to the above method.

[0130] At this point, the optimal control input sequence of the air-conditioning system under specific environmental parameters in various dimensions within a certain period of time in the future is obtained, so that the environmental parameters in various dimensions of the room meet the target environmental parameter range, that is, meet the requirements of a clean environment while ensuring minimum energy consumption.

[0131] Step S004: Control the air conditioning system based on the optimal control input sequence of each dimension.

[0132] It should be noted that, based on the optimal control input sequence obtained in each dimension, the cleanliness of the current indoor environment can be achieved with the minimum energy consumption of the air-conditioning system.

[0133] 1) Initialize LSTM model training parameters; set LSTM weights and paranoia ;

[0134] 2) Initialize the MPC parameters and set them to the data set of the first moment of the air conditioning system ;

[0135] 3) Initialize energy consumption weights and environmental parameter weights;

[0136] 4) Obtain the initial state of the air conditioner and initial environmental parameters;

[0137] 5) Data preprocessing: cleaning, interpolation and denoising of data;

[0138] 6) LSTM prediction: Use the preprocessed data to train the neural network model to obtain the power change trend of each dimension and the predicted values ​​of environmental parameters in the future;

[0139] 7) MPC optimization: obtaining the optimal control input sequence ;

[0140] 8) Control input implementation: exist Input the air conditioning system within the time period to control the air conditioning;

[0141] 9) System status update: collect the latest environmental parameters of each dimension in real time and update the objective function;

[0142] 10) Feedback adjustment: According to the prediction error, adjust the MPC optimization parameters.

[0143] At this point, the environmental parameters of each dimension are predicted through the LSTM neural network model, the power change trend of each dimension is constructed, and then the energy consumption of the air-conditioning system in the future is analyzed. Combined with the difference between the predicted values ​​of the environmental parameters in the future and the requirements of the clean environment, the optimal control input sequence of each dimension is obtained by optimization, so as to control the air-conditioning system and finally achieve the minimum energy consumption of the air-conditioning system while adjusting the clean environment.

[0144] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A clean air conditioning control method based on neural network and MPC algorithm, characterized in that: The method comprises the following steps: Collect multi-dimensional environmental data indoors and outdoors, and obtain multi-dimensional status data and operation data of the air-conditioning system; Construct an LSTM neural network model, input the environmental parameters of each dimension at each moment, and output the predicted values ​​of the environmental parameters of each dimension at the next moment. Based on the predicted values ​​of the environmental parameters and the environmental parameters at adjacent moments, the power change trend of the air-conditioning system is obtained by training the LSTM neural network model. The target environmental parameter range of each dimension is set based on the indoor clean conditions. Combined with the control input sequence and power change trend of each dimension, and the predicted value of each dimension of environmental parameters output by the LSTM neural network model, the energy consumption cost function of each dimension in the future time is constructed, and then the optimal control input sequence of each dimension is obtained by optimization. Control the air conditioning system based on the optimal control input sequence in each dimension; The specific method of constructing the LSTM neural network model includes: Define an N-layer LSTM neural network model, with M neural units in each layer; Initialize network weights W and bias vector b; Construct a loss function based on model parameters according to the input and output of the LSTM neural network model; Use the back-propagation algorithm to calculate the gradient of the loss function with respect to each weight; Apply gradient descent method to update model parameters; Train iterative LSTM neural network model; The specific method of obtaining the predicted values ​​of the environmental parameters of each dimension at the next moment includes: According to the data set X at the tth moment t Construct the input feature vector Y t , the data set X t Including environmental parameters of each dimension at the tth moment; Output vector Y' t+1 , based on the output vector, we get the predicted data set X' at the t+1th moment t+1 , the prediction data set X′ t+1 Including the predicted values ​​of environmental parameters in each dimension at the t+1th moment; The specific method of obtaining the power change trend of the air conditioning system includes: X′ t+1 ×Y′ t+1 =LSTM(X t ,u t ) Among them, X t represents the input feature vector at the tth moment, X' t+1 Represents the predicted data set at time t+1, Y' t+1 represents the output vector at the t+1th moment, u t represents the control input at the tth moment; The control inputs at all times constitute the control input sequence U, and then the power change trend f(u) of the corresponding air-conditioning system is obtained; The specific method of constructing the energy consumption cost function in each dimension in the future time includes: Set the future time and obtain the predicted data set for each moment in the future based on the LSTM neural network model; based on the power change trend of the air-conditioning system and the predicted data set for each moment in the future, construct the energy consumption cost function as follows: Where R represents the energy consumption cost function of air conditioning; f(u T+Δt ) represents the energy consumption of the air conditioner when the control sequence U is input using the power change trend function f(u) at the T+Δtth moment; H T+Δt represents the predicted indoor environmental parameters at the T+Δtth moment; H represents the reference parameter of the environmental parameters in this dimension; ω e With ω h represents the weight factor; U = {u T ,u T+1 ,u T+2 ,…,u T+Δt ,…,u T+ΔT } is the control input sequence of the air conditioner.

2. The clean air conditioning control method based on neural network and MPC algorithm according to claim 1 is characterized in that: The loss function based on the model parameters is constructed according to the input and output of the LSTM neural network model, including the following specific methods: Where N is the number of samples in the training set; θ is the model parameter; Y t+1 represents the input feature vector at the t+1th moment; L() represents the loss function.

3. The clean air conditioning control method based on neural network and MPC algorithm according to claim 1 is characterized in that: The specific method of setting the target environmental parameter range of each dimension based on the indoor cleanliness condition is as follows: For any dimension of environmental parameters, set the target environmental parameter range [H min ,H max ].

4. The clean air conditioning control method based on neural network and MPC algorithm according to claim 1 is characterized in that: The specific method of setting the future time and obtaining the predicted data set for each moment in the future time based on the LSTM neural network model includes: Through the LSTM neural network model, the prediction data set for each moment in the future ΔT time is obtained.

5. The clean air conditioning control method based on neural network and MPC algorithm according to claim 1 is characterized in that: The specific method for obtaining the weight factor is as follows: Energy consumption weight ω e =1-(x+Δx); environmental parameter weight ω h =x+Δx; where x is the weight adjustment parameter and Δx is the weight adjustment factor; Calculate energy consumption error and environmental parameter error at each moment in the future; The weight adjustment factor is adjusted according to the energy consumption error and environmental parameter error. The specific calculation process is: Among them, Δx represents the weight adjustment factor, ΔT represents the number of future moments, and They represent the energy consumption error and environmental parameter error at the T+Δtth moment respectively, and sigmoid represents the sigmoid function.

6. The clean air conditioning control method based on neural network and MPC algorithm according to claim 5 is characterized in that: The specific method of obtaining the optimal control input sequence of each dimension includes: Energy consumption error where f'(u T+Δt ) represents the actual energy consumption of the LSTM model of the air-conditioning system at the T+Δtth moment; Environmental parameter error Where H' T+Δt Indicates the actual value of the environmental parameter of the LSTM model at the T+Δtth moment.

Images