A Predictive Optimization Control Algorithm and Device for a Constant Temperature and Humidity Air Conditioning System

Abstract

The present invention provides a predictive optimization control algorithm and device for a constant temperature and humidity air conditioning system, including: a control module, an air handling unit, a temperature and humidity sensor, a solar radiation sensor, a water valve opening collector, a fan frequency collector, a first water valve controller, a second water valve controller, a fan frequency converter, and a primary filter; the control module includes a storage unit, a CPU, a signal input interface, and an RS485 communication interface; a predictive optimization control algorithm program for a constant temperature and humidity air conditioning system is stored in the control module. This algorithm realizes intelligent optimization by means of mutual learning within the country and mutual learning between countries. By selecting excellent learning objects, the samples can learn from the better group, avoiding the problem of falling into local optima. The present invention can achieve high-precision control of the temperature and humidity of the constant temperature and humidity air conditioning system, and at the same time achieve the maximum energy-saving effect of the constant temperature and humidity air conditioning system.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and specifically to a prediction optimization control algorithm and device for a constant temperature and humidity air conditioning system. Background Art

[0002] In actual production and life, many places have very high requirements for the temperature and humidity of the environment, and need to strictly control the fluctuation range of temperature and humidity. Constant temperature and humidity air conditioning systems are widely used in such places.

[0003] In traditional control methods, constant temperature and humidity air conditioning systems usually operate at a constant air volume. In a constant air volume system, the air supply volume is calculated based on the maximum load. When the system is in a partial load condition, it will inevitably lead to too low supply air temperature. Therefore, it is necessary to additionally heat the air, resulting in unnecessary energy consumption. Although a variable air volume system can overcome the above problems, for a constant temperature and humidity air conditioning system with strong nonlinearity and high temperature and humidity coupling, only using traditional control methods cannot meet the indoor temperature and humidity requirements. Therefore, it is necessary to find a better control method to meet the requirements of a variable air volume constant temperature and humidity air conditioning system.

[0004] Currently, the commonly used control methods include PID control and intelligent control. Intelligent control includes neural network control, genetic algorithm control, model predictive control, etc. Using the PID control strategy, the PID controller compares the deviation between the indoor temperature and humidity and the target set value, and outputs a control signal to adjust the chilled water valve, hot water valve and humidifying valve, so that the indoor temperature and humidity are maintained within the set range. It has hysteresis and cannot establish a system model, and it is difficult to meet higher control requirements. Using a neural network model, the indoor temperature and humidity at future moments are predicted through known data, and then different air conditioning operation controls are traversed to select the optimal control parameters to meet the indoor temperature and humidity requirements. A system model is established with a neural network to pre-control the indoor temperature and humidity in advance, but the control effect is not good.

[0005] Model predictive control, on the basis of establishing a system model for prediction, incorporates a rolling optimizer, and combines an optimization algorithm to achieve optimal control of the system. Currently, the optimization algorithms used in model predictive control usually have problems such as low convergence accuracy, poor stability, and being easily trapped in local optima, and need to be further improved and optimized. Summary of the Invention

[0006] (1) Technical Problems to be Solved

[0007] Aiming at the deficiencies of the prior art, the present invention provides a prediction optimization control algorithm and device for a constant temperature and humidity air conditioning system, and solves the problem of poor prediction optimization control effect of the constant temperature and humidity air conditioning system proposed in the above background art.

[0008] (2) Technical Solutions

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0010] The present invention provides a prediction and optimization control device for a constant temperature and humidity air conditioning system, including: a control module, an air handling unit, temperature and humidity sensors, a solar radiation sensor, a water valve opening collector, a fan frequency collector, a first water valve controller, a second water valve controller, a fan frequency converter, and a primary filter; wherein, the control module includes a storage unit, a CPU, a signal input interface, and an RS485 communication interface; the air handling unit includes: a surface cooler, a reheater, a variable frequency fan, and a primary filter.

[0011] A prediction and optimization control algorithm program for a constant temperature and humidity air conditioning system is stored in the control module; temperature and humidity sensors and a solar radiation sensor are placed outdoors to collect the outdoor air dry bulb temperature, outdoor air relative humidity, and solar radiation irradiance; temperature and humidity sensors are placed at the air supply outlet to collect the air supply dry bulb temperature and air supply relative humidity; temperature and humidity sensors are placed indoors to collect the indoor air dry bulb temperature and indoor air relative humidity; a water valve opening collector is placed at the coil water valves of the surface cooler and the reheater to obtain the cold coil water valve opening and the hot coil water valve opening; a fan frequency collector is placed in the fan frequency converter to obtain the fan frequency; all the collected data is transmitted into the storage unit through the signal input interface; after the optimization calculation by the control module, the control signal is transmitted to the first water valve controller and the second water valve controller through the RS485 communication interface, the first water valve controller adjusts the cold coil water valve opening, the second water valve controller adjusts the hot coil water valve opening, and the fan frequency converter adjusts the fan frequency.

[0012] The present invention also provides a prediction and optimization control algorithm for a constant temperature and humidity air conditioning system, which controls a prediction and optimization control device for a constant temperature and humidity air conditioning system as described above, including:

[0013] Collect and obtain data, input the collected data into the neural network model, and train the neural network system prediction model;

[0014] Design a rolling optimizer, determine the prediction domain and control domain of the system prediction model controller, and set the objective function;

[0015] The control module obtains the prediction model input parameters during the operation of the system through each sensor, inputs them into the neural network system prediction model, adjusts the control parameters through an intelligent optimization algorithm based on the national evolution process, minimizes the objective function, and outputs the optimal control parameters;

[0016] The control module transmits the output optimal control parameters to the chilled water valve controller of the finned coil heat exchanger, the reheater water valve controller, and the fan frequency converter in the form of digital signals, so as to adjust the fan frequency, the chilled water flow rate, and the hot water flow rate, and change the air supply volume, the air supply temperature, and the air supply humidity.

[0017] Preferably, the input parameters of the prediction model include: the indoor air temperature T z (k) at the current moment, the indoor air moisture content W z (k), the air supply temperature T sa (k), the air supply moisture content W sa (k), the room cooling load and the room moisture load the fan frequency f(k - 1) at the previous moment, the opening degree L cc (k - 1) of the cold coil water valve, the opening degree L hc (k - 1) of the hot coil water valve.

[0018] Preferably, the data collected and obtained include: the outdoor air temperature and moisture content, the air supply temperature and moisture content, the indoor air temperature and moisture content, the fan frequency, the opening degree of the cold coil water valve, the opening degree of the hot coil water valve, the room cooling load, and the room moisture load;

[0019] The room cooling load and the room moisture load in the obtained data can be calculated by the following formula, where the solar irradiance is collected by a solar radiation sensor;

[0020]

[0021]

[0022]

[0023] In the formula: represents the room cooling load, W; G sa represents the air supply volume, kg / s; c a represents the specific heat capacity of air, J / (kg·℃); T sa represents the air supply temperature, ℃; k wall represents the heat transfer coefficient of the wall, W / (m 2 ·℃); A wall represents the area of the wall, m 2 ; T oa represents the outdoor air temperature, ℃; A e represents the effective area of the building, m 2 ; I represents the solar irradiance, W / m 2 ; represents the cooling load formed by the people in the room per unit time, W; Represents the cooling load formed by other factors in the room, such as lamps, equipment, etc., W; W represents the moisture content, kg / (kg dryair); Represents the relative humidity, %; P q,b Represents the saturated water vapor pressure, Pa; B represents the atmospheric pressure, Pa; Represents the room moisture load, kg / s; W sa Represents the moisture content of the supply air, kg / (kgdryair); Represents the moisture load of the people in the room, kg / s;

[0024] During the training process of the neural network, the mean square error (MSE) is used to evaluate the training effect:

[0025]

[0026] In the formula: Represents the target output parameter; Represents the actual output parameter; N represents the number of samples.

[0027] Preferably, the design rolling optimizer determines the prediction domain and control domain of the system prediction model controller, and sets the objective function, including:

[0028] The objective function consists of the tracking errors of the predicted indoor air temperature and indoor air moisture content, and the equipment energy consumption. Among them, the equipment energy consumption consists of the fan energy consumption, the chiller power consumption, and the reheater power consumption;

[0029]

[0030]

[0031]

[0032] F3 = P fan +P cc +P hc (8)

[0033] f min <f<f max (9)

[0034] L cc,min <L cc <L cc,max (10)

[0035] L hc,min <L hc <L hc,max (11)

[0036] In the formula: N p Represents the prediction domain; wi Denote the weights, where \(i = 1, 2, 3\); \(F_1\) represents the tracking error of predicting the indoor air temperature; \(F_2\) represents the tracking error of predicting the indoor air moisture content. Denote the indoor air temperature predicted by the model, in °C; \(T\) z,r (k + i) denotes the reference trajectory of the indoor air temperature, in °C; Denote the indoor air moisture content predicted by the model, in kg / (kg dry air); \(W\) z,r (k + i) denotes the reference trajectory of the indoor air moisture content, in kg / (kg dry air); The subscript "fan" represents the fan, "cc" represents the cooling coil, "hc" represents the heating coil, "min" represents the minimum value, "max" represents the maximum value; \(P\) represents the equipment energy consumption, in W; \(f\) represents the fan frequency, in Hz; \(L\) represents the opening degree of the water valve, in %.

[0037] For the constant temperature and humidity air conditioning system, the fresh air mode is adopted, and there are linear relationships between the air supply volume and the fan frequency, and between the water flow and the opening degree of the water valve.

[0038] In the objective function, the equivalent energy consumption of the cooling coil is expressed by the following formula, that is, by adjusting the opening degree of the chilled water valve, the chilled water flow is adjusted, thereby changing the enthalpy value of the air after being processed by the cooling coil, and the equivalent energy consumption of the cooling coil is adjusted.

[0039]

[0040] s.t. \(T\) ca = \(f\) coil (T oa , \(W\) oa , \(G\) sa , \(G\) cw )

[0041] = \(T\) oa + \(G\) cw [\(β_1T\) oa +\(β_2W\) oa +\(β_3G\) sa +\(β_4G\) cw +\(β_5 + β_6T\) oa 2 +

[0042] \(β_7W\) oa 2 ++\(β_8G\) sa 2 +\(β_9G\) cw 2 +\(β\) 10 \(T\) oa \(W\) oa +\(β\) 11 \(W\) oa \(G\) sa +

[0043] β 12 G sa G cw +β 13 G cw T oa +β 14 T oa G sa +β 15 W oa G cw (13)

[0044] W ca =g coil (T oa ,W oa ,G sa ,G cw )

[0045] =W oa +G cw [γ1T oa +γ2W oa +γ3G sa +γ4G cw +γ5+γ6T oa 2 +

[0046] γ7W oa 2 ++γ8G sa 2 +γ9G cw 2 +γ 10 T oa W oa +γ 11 W oa G sa +

[0047] γ 12 G sa G cw +γ 13 G cw T oa +γ 14 T oa G sa +γ 15 W oa G cw (14)

[0048] h ca =1.01T ca +W ca (2501+1.84T ca ) (15)

[0049]

[0050] In the formula: the subscript cc represents the cold coil, sa represents the supply air, oa represents the outdoor air, ca represents the air after being treated by the cold coil, and cw represents the chilled water passing through the cold coil; h oa represents the enthalpy value of the outdoor air. When the operating condition is the return air condition, this item is the enthalpy value of the air after mixing the fresh air and the return air, J; h ca represents the enthalpy value of the air after being treated by the cold coil, J; T ca represents the temperature of the air after being treated by the cold coil, °C; W ca represents the moisture content of the air after being treated by the cold coil, kg / (kg dry air); W oa represents the moisture content of the outdoor air, kg / (kg dry air); η cc represents the efficiency of the cold coil; COP c represents the performance coefficient of the chiller; G cw represents the chilled water flow rate passing through the cold coil, kg / s; G cw,min and G cw,max represent the minimum and maximum values of the chilled water flow rate passing through the cold coil, kg / s; β i (i = 0, 1, …, 15), γ i (i = 0, 1, …, 15) respectively represent the fitting coefficients of the temperature and humidity of the air after being treated by the cold coil. These coefficients can be obtained through simulation software or experimental data;

[0051] In the objective function, the equivalent energy consumption of the hot coil is expressed by the following formula, that is, by adjusting the opening of the hot water valve and the hot water flow rate, the temperature of the air after being treated by the hot coil is changed to adjust the equivalent energy consumption of the hot coil:

[0052]

[0053] s.t. T sa = f coil (T ca , W ca , G sa , G hw )

[0054] = T ca + G hw [β1T ca + β2W ca + β3G sa + β4G hw + β5 + β6T ca 2 +

[0055] β7W ca2 ++β8G sa 2 +β9G hw 2 +β 10 T ca W ca +β 11 W ca G sa +

[0056] β 12 G sa G hw +β 13 G hw T ca +β 14 T ca G sa +β 15 W ca G hw (18)

[0057]

[0058] In the formula: the subscript hc represents the heat coil, and hw represents the hot water passing through the heat coil; η hc represents the efficiency of the heat coil; COP h represents the performance coefficient of the boiler; G hw represents the hot water flow rate passing through the heat coil, kg / s; G hw,min and G hw,max represent the maximum value of the hot water flow rate passing through the heat coil, kg / s;

[0059] The objective function (5) can be normalized to obtain the following formula:

[0060]

[0061] In the formula: F min and F max are the minimum and maximum values of the objective function respectively;

[0062] In the model predictive control, the reference trajectory is obtained by the following formula:

[0063]

[0064] In the formula: represents the set values of the indoor temperature and moisture content; α is the softening coefficient, 0 < α < 1.

[0065] Preferably, adjusting the control parameters and minimizing the objective function through the intelligent optimization algorithm based on the national evolution process, and outputting the optimal control parameters, includes:

[0066] S1: Initialize the country

[0067] Determine the policy parameters for the initially established country and set the number of policy parameters to D. Use Equation (21) as the objective function to calculate the fitness value of the initial policy parameters; The parameters of the initial policy can be expressed as:

[0068] In the formula: x

[0069]

[0070] represents the value of the d-th dimension in the policy parameters; d

[0071] S2: The country determines the preliminary policy adjustment

[0072] Randomly generate a random number R uniformly distributed between (0, 1). If R > 0.8, make a large adjustment; otherwise, make a small adjustment. After adjusting the policy parameters, calculate the fitness value of the new parameters;

[0073] Large adjustment: Randomly generate new parameters using Levy flight. The method is as follows:

[0074]

[0075]

[0076] In the formula: u ~ N(0, σ 2 ) and v ~ N(0, 1). Usually, β = 1.5;

[0077]

[0078]

[0079] Calculate the new parameters as follows: If x′ d exceeds the parameter range, then:

[0080] x″ d = |x′ d - x′ d,int |·(x d,max - x d,min ) + x d,min (26)

[0081] In the formula: x″ d is the new parameter of the d-th dimension, x′ d,int is the integer part of x′ d , x d,max and x d,min are the upper and lower limits of the d-th dimension data range respectively;

[0082] ​Minor adjustment: In this embodiment, the parameters of the large country corresponding to the dimension are used as the benchmark, and a random number that satisfies the normal distribution of x~N(0,1) is added as the new parameter, that is:

[0083] x′ d =x d +x (27)

[0084] In the formula: x′ d is the new parameter of the d-th dimension; x d is the original parameter of the d-th dimension; x~N(0,1);

[0085] If x′ d exceeds the parameter range, then there is:

[0086]

[0087] In the formula, x″ is the new parameter of the d-th dimension, x d,max and x d,min are the upper and lower limits of the range of the d-th dimension data respectively;

[0088] S3: Determine whether the fitness of the policy parameters after the preliminary adjustment has improved

[0089] If the fitness of the parameters improves after the preliminary adjustment, execute step S2 and continue the policy adjustment; otherwise, execute step S4;

[0090] S4: The country updates the policy parameters

[0091] The country takes the parameter with the best fitness in the preliminary policy adjustment as the new national policy;

[0092] S5: The country splits into small countries

[0093] The large country splits into n small countries. Each small country randomly retains some of the parameters of the large country, and the remaining parameters are slightly adjusted in this embodiment in the same way as the minor adjustment method in S2. The policy parameters established by the small countries can be expressed as:

[0094] X i =(x i,1 ,…,x i,d ,…,x i,D ) (29)

[0095] s.t.d∈[1,D],i∈[1,n]

[0096] In the formula: x i,d represents the policy parameter value of the d-th dimension of the i-th small country;

[0097] If the fitness of the newly generated parameters becomes worse compared with the fitness of the large country, the parameters of the small countries are not adjusted, and all the parameters of the large country are retained.

[0098] S6: Each small country assigns officials to different positions

[0099] Each small country sets up m officials, among which 40% are ordinary court officials, 40% are spies, and 20% are prime ministers;

[0100] Define the responsibilities for each official: Ordinary court officials put forward suggestions and make policy adjustments. Spies randomly go to other small countries to learn better policies from the court officials or prime ministers of other countries. The prime minister widely listens to the suggestions of the court officials and spies in his own country, learns from the better policies among the court officials and spies, and finally obtains the best policy parameters as the new policy parameters of the small country;

[0101] Each official needs to adjust the policy parameters of each small country, that is, the initial parameters of each official are as shown in formula (29);

[0102] S7: Court officials put forward suggestions and make policy adjustments

[0103] Each court official in each small country starts to put forward suggestions and adjust the policies of their own small country. Each court official generates a random number R uniformly distributed between (0, 1). If R > 0.8, a large adjustment is made, otherwise a small adjustment is made, and the fitness value of the parameters after each adjustment is calculated;

[0104] The large adjustment method uses Levy flight to generate new parameters, and the small adjustment method uses the normal distribution to generate new random numbers. The calculation steps are the same as S2. Each court official makes c policy adjustments and records the parameters with the best fitness in the policy adjustments of each court official;

[0105] S8: Spies go to other countries to learn for the first time

[0106] Each spy in each small country randomly selects another small country and learns the policy parameters proposed by the court officials of other countries in a certain way, and calculates the fitness value of the parameters after learning;

[0107] The PSO (Particle Swarm Optimization) algorithm is used for learning. The spy selects the best parameters among the court officials of other countries as the learning object. That is, when initializing the particles in the PSO algorithm, each particle has an 80% probability of taking the learning object parameters as the initial position of the particle and a 20% probability of randomly initializing the position, and then performs PSO iterative calculation. The optimized best particle position is output as the new policy parameters that can be adopted after the spy's learning. Compare the fitness of this parameter with the fitness of the spy's original parameters. If it becomes worse, the spy does not adopt the suggestions of the court officials of other countries and the policy parameters remain unchanged;

[0108] The above process is repeated s times, that is, the spy randomly selects s small countries for policy learning, and records the set of best parameters after the first learning of the spies in each small country;

[0109] S9: The prime minister collects the opinions of all officials for the first time

[0110] Each prime minister of each small country selects the top 30% of the best policy parameters among the courtiers and spies in their respective small countries as the objects of learning, and learns from them and adjusts the initial policy parameters of the small country in a certain way;

[0111] Use the same PSO particle swarm algorithm as in step S8 for learning. Each prime minister randomly selects a certain parameter from the top 30% of the best policy parameters among the courtiers and spies in their own country as the object of learning, performs PSO iterative calculation, and outputs the optimized best particle position as the new policy parameter after the prime minister's learning. This process is repeated p times, and the best parameter obtained during the p learning processes is used as the new parameter of the prime minister. Record the set of the best parameters for the initial learning of the prime ministers of each small country.

[0112] S10: Spies go to other countries to learn for the second time

[0113] Each spy in each small country randomly selects another small country, learns from the policy parameters of the prime minister of the other country in a certain way, and calculates the fitness value of the parameters after learning;

[0114] Use the PSO particle swarm algorithm for learning. The specific steps are the same as in S8. The difference is that the spy randomly selects the best parameter among the prime ministers of other countries as the object of learning. Record the set of the best parameters for the second learning of the spies in each small country;

[0115] S11: Prime ministers collect the opinions of all officials for the second time

[0116] Each prime minister selects the set of the top 30% of the best policy parameters among the courtiers and spies in their respective small countries, learns from them and adjusts the policy parameters in a certain way;

[0117] Use the PSO particle swarm algorithm for learning. The specific steps are the same as in S9. Record the set of the best parameters for the second learning of the prime ministers of each small country;

[0118] S12: Obtain the best policy parameters among all small countries

[0119] Each small country updates its policy parameters: By comparing the fitness values of the parameters through the sets of the best parameters obtained from the initial and second learning of the prime ministers of each small country, obtain the best policy parameter among the prime ministers of each small country, and use it as the new policy parameter of each small country;

[0120] Compare the new policy parameters of each small country to obtain the parameter with the best fitness among all small countries, which can be expressed as:

[0121] X best =(x best,1 ,…,x best,d ,…,x best,D ) (30)

[0122] s.t.d∈[1,D]

[0123] where: x best,d represents the value of the d-th dimension of the optimal policy parameter among all small countries;

[0124] S13: The small country with the optimal policy parameter unifies all countries

[0125] The small country with the optimal policy parameter unifies all countries and becomes a new big country, taking its policy parameter as the initial policy parameter of the new country;

[0126] S14: Determine whether the iteration termination condition is reached

[0127] The iteration termination condition of this algorithm is to reach the maximum number of iterations or reach the specified convergence accuracy. If the iteration termination condition is reached, step S15 is executed; otherwise, S2 is executed.

[0128] S15: Output the optimal policy parameter

[0129] After continuous evolution of the country, finally the big country with the optimal policy parameter is output as the optimal value to complete the optimization calculation.

[0130] (III) Beneficial effects

[0131] The present invention provides a predictive optimization control algorithm and device for a constant temperature and humidity air conditioning system, having the following beneficial effects:

[0132] The predictive optimization control algorithm and device for a constant temperature and humidity air conditioning system provided by the present invention can achieve high-precision control of temperature and humidity, while reducing the system energy consumption and realizing energy-saving optimization. The intelligent optimization algorithm based on the national evolution process uses the way of mutual learning among small countries and mutual learning between countries, enabling the algorithm to have good global search ability in the early stage of iteration; by selecting excellent learning objects, the samples can learn from the better group, avoiding the problem of falling into local optimum; it has the ability to accommodate various existing intelligent optimization algorithms, and can select a suitable algorithm according to the actual problem, allowing the prime minister and the spies to approach the excellent learning objects with the existing intelligent optimization algorithms, and can achieve fast convergence with fewer loop times, solving the problem of poor predictive optimization control effect in the prior art for constant temperature and humidity systems. Description of the drawings

[0133] Figure 1 is a structural diagram of a predictive optimization control device for a constant temperature and humidity air conditioning system provided by an embodiment of the present invention;

[0134] Figure 2 is a schematic diagram of a neural network structure in a predictive optimization control algorithm and device for a constant temperature and humidity air conditioning system provided by an embodiment of the present invention;

[0135] Figure 3 Schematic diagram of the prediction and optimization control process of a constant temperature and humidity air conditioning system provided by an embodiment of the present invention;

[0136] Figure 4 Schematic diagram of the prediction and optimization control algorithm flow of a constant temperature and humidity air conditioning system provided by an embodiment of the present invention

[0137] Figure 5 Schematic diagram of the intelligent optimization algorithm flow based on the national evolution process in the prediction and optimization control algorithm of a constant temperature and humidity air conditioning system provided by an embodiment of the present invention;

[0138] Figure 6 Convergence curve graph of the prediction and optimization control algorithm - CEA and PSO, SBO, SSA algorithms of a constant temperature and humidity air conditioning system based on the national evolution process provided by an embodiment of the present invention under different test functions;

[0139] Figure 7 and Figure 8 is the change graph of the indoor temperature and relative humidity under the prediction and optimization control of a constant temperature and humidity air conditioning system proposed by an embodiment of the present invention;

[0140] Figure 9 is the power graph of the prediction and optimization control system of a constant temperature and humidity air conditioning system proposed by an embodiment of the present invention;

[0141] In the figure:

[0142] Control module 1, storage unit 2, CPU 3, signal input interface 4, RS485 communication interface 5, air handling unit 6, surface cooler 7, reheater 8, variable frequency fan 9, temperature and humidity sensor 10, solar radiation sensor 11, water valve opening collector 12, fan frequency collector 13, first water valve controller 14, second water valve controller 15, fan frequency converter 16, primary filter 17. Specific implementation manner

[0143] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention.

[0144] As Figure 1 shown, an embodiment of the present invention provides a prediction and optimization control device for a constant temperature and humidity air conditioning system, including: control module 1, air handling unit 6, temperature and humidity sensor 10, solar radiation sensor 11, water valve opening collector 12, fan frequency collector 13, first water valve controller 14, second water valve controller 15, fan frequency converter 16, primary filter 17;

[0145] Among them, the control module 1 includes a storage unit 2, a CPU 3, a signal input interface 4, and an RS485 communication interface 5; the air handling unit 6 includes: a surface cooler 7, a reheater 8, a variable-frequency fan 9, and a primary filter 17;

[0146] A constant temperature and humidity air conditioning system prediction and optimization control algorithm program is stored in the control module 1; a temperature and humidity sensor 10 and a solar radiation sensor 11 are placed outdoors to collect the outdoor air dry bulb temperature, outdoor air relative humidity, and solar radiation irradiance; a temperature and humidity sensor 10 is placed at the air supply outlet to collect the air supply dry bulb temperature and air supply relative humidity; a temperature and humidity sensor 10 is placed indoors to collect the indoor air dry bulb temperature and indoor air relative humidity; a valve opening collector 12 is placed at the coil water valves of the surface cooler and the reheater to obtain the cold coil water valve opening and the hot coil water valve opening; a fan frequency collector 13 is placed in the fan frequency converter to obtain the fan frequency; all the collected data is transmitted into the storage unit 2 through the signal input interface 4;

[0147] After the optimization calculation by the control module 1, the control signal is transmitted to the first water valve controller 14 and the second water valve controller 15 through the RS485 communication interface 5. The first water valve controller 14 adjusts the cold coil water valve opening, and the second water valve controller 15 adjusts the hot coil water valve opening. The fan frequency is adjusted by the fan frequency converter 16.

[0148] As Figures 2 - 4 shown, the present invention also provides a constant temperature and humidity air conditioning system prediction and optimization control algorithm for controlling a constant temperature and humidity air conditioning system prediction and optimization control device as described above, including:

[0149] 401 Collect and obtain data, input the collected data into the neural network model, and train the neural network system prediction model;

[0150] 402 Design a rolling optimizer, determine the prediction domain and control domain of the system prediction model controller, and set the objective function;

[0151] 403 The control module obtains the prediction model input parameters during the operation of the system through each sensor, inputs them into the neural network system prediction model, adjusts the control parameters through the intelligent optimization algorithm based on the national evolution process, minimizes the objective function, and outputs the optimal control parameters;

[0152] 404 The control module transmits the output optimal control parameters to the surface cooler water valve controller, the reheater water valve controller, and the fan frequency converter in the form of digital signals, so as to realize the adjustment of the fan frequency and the cold water flow rate and the hot water flow rate, and change the air supply volume, air supply temperature, and air supply humidity.

[0153] Preferably, the prediction model input parameters include: the indoor air temperature T at the current moment z(k), indoor air moisture content W z (k), supply air temperature T sa (k), supply air moisture content W sa (k), room cooling load and room moisture load The fan frequency f(k - 1) and the cold coil water valve opening L at the previous moment cc (k - 1), the hot coil water valve opening L hc (k - 1).

[0154] Preferably, the state parameters actually output by the system are representing the actual indoor air temperature and moisture content of the system at time k; the model prediction output parameters are representing the indoor air temperature and moisture content predicted by the model at time k + i; the reference trajectory of the indoor air temperature and moisture content at time k + i is The optimal control parameters at time k + j are The disturbance parameter of the system at time k is respectively representing the cooling load and moisture load indoors.

[0155] Preferably, the data collected and obtained includes: outdoor air temperature and moisture content, supply air temperature and moisture content, indoor air temperature and moisture content, fan frequency, cold coil water valve opening, hot coil water valve opening, room cooling load, room moisture load;

[0156] The room cooling load and room moisture load in the obtained data can be calculated by the following formula, where the solar radiation illuminance is collected by a solar radiation sensor;

[0157]

[0158]

[0159]

[0160] In the formula: represents the room cooling load, W; G sa represents the supply air volume, kg / s; c a represents the specific heat capacity of air, J / (kg·℃); T sa represents the supply air temperature, ℃; k wall represents the heat transfer coefficient of the wall, W / (m 2 ·℃); A wall represents the area of the wall, m 2 ; T oa represents the outdoor air temperature, ℃; A e represents the effective area of the building, m 2 ; I represents the solar radiation illuminance, W / m 2; Represents the cooling load formed by the personnel in the room per unit time, in W; Represents the cooling load formed by other factors in the room, such as lamps, equipment, etc., in W; W represents the moisture content, in kg / (kg dryair); Represents the relative humidity, in %; P q,b Represents the saturated water vapor pressure, in Pa; B represents the atmospheric pressure, in Pa; Represents the moisture load of the room, in kg / s; W sa Represents the moisture content of the supply air, in kg / (kgdryair); Represents the moisture load of the personnel in the room, in kg / s;

[0161] During the training process of the neural network, the mean squared error (MSE) is used to evaluate the training effect:

[0162]

[0163] In the formula: Represents the target output parameter; Represents the actual output parameter; N represents the number of samples.

[0164] Preferably, the designed rolling optimizer determines the prediction domain and control domain of the system prediction model controller, and sets the objective function, including:

[0165] The objective function consists of the tracking errors of the predicted indoor air temperature and indoor air moisture content, and the equipment energy consumption. Among them, the equipment energy consumption consists of the fan energy consumption, the electric power consumption of the surface cooler, and the electric power consumption of the reheater;

[0166]

[0167]

[0168]

[0169] F3 = P fan + P cc + P hc (38)

[0170] f min <f <f max (39)

[0171] L cc,min <L cc <L cc,max (40)

[0172] L hc,min <L hc <L hc,max (41)

[0173] In the formula: N p represents the prediction domain; w i represents the weight, where i = 1, 2, 3; F1 represents the tracking error of predicting the indoor air temperature; F2 represents the tracking error of predicting the indoor air moisture content; represents the indoor air temperature predicted by the model, °C; T z,r (k + i) represents the reference trajectory of the indoor air temperature, °C; represents the indoor air moisture content predicted by the model, kg / (kgdryair); W z,r (k + i) represents the reference trajectory of the indoor air moisture content, kg / (kg dryair); the subscript fan represents the fan, cc represents the cooling coil, hc represents the heating coil, min represents the minimum value, max represents the maximum value; P represents the equipment energy consumption, W; f represents the fan frequency, Hz; L represents the opening degree of the water valve, %;

[0174] For the constant temperature and humidity air conditioning system, the fresh air mode is adopted, and there are linear relationships between the air supply volume and the fan frequency, and between the water flow and the opening degree of the water valve;

[0175] In the objective function, the equivalent energy consumption of the cooling coil is expressed by the following formula, that is, by adjusting the opening degree of the chilled water valve, the chilled water flow is adjusted, thereby changing the enthalpy value of the air after being processed by the cooling coil, and the equivalent energy consumption of the cooling coil is adjusted;

[0176]

[0177] s.t. T ca = f coil (T oa , W oa , G sa , G cw )

[0178] = T oa + G cw [β1T oa + β2W oa + β3G sa + β4G cw + β5 + β6T oa 2 + (43)

[0179] β7W oa 2 ++ β8G sa 2 + β9G cw 2 + β 10 T oa W oa + β11 W oa G sa +

[0180] β 12 G sa G cw +β 13 G cw T oa +β 14 T oa G sa +β 15 W oa G cw

[0181] W ca =g coil (T oa ,W oa ,G sa ,G cw )

[0182] =W oa +G cw [γ1T oa +γ2W oa +γ3G sa +γ4G cw +γ5+γ6T oa 2 + (44)

[0183] γ7W oa 2 ++γ8G sa 2 +γ9G cw 2 +γ 10 T oa W oa +γ 11 W oa G sa +

[0184] γ 12 G sa G cw +γ 13 G cw T oa +γ 14 T oa G sa +γ 15 W oa G cw

[0185] h ca =1.01T ca +W ca ​​(2501 + 1.84T ca ) (45)

[0186]

[0187] Where: the subscript cc represents the cold coil, sa represents the supply air, oa represents the outdoor air, ca represents the air after being treated by the cold coil, cw represents the chilled water passing through the cold coil; h oa represents the enthalpy value of the outdoor air. When the operating condition is the return air condition, this item is the enthalpy value of the mixed air of the fresh air and the return air, J; h ca represents the enthalpy value of the air after being treated by the cold coil, J; T ca represents the temperature of the air after being treated by the cold coil, °C; W ca represents the moisture content of the air after being treated by the cold coil, kg / (kg dry air); W oa represents the moisture content of the outdoor air, kg / (kg dry air); η cc represents the efficiency of the cold coil; COP c represents the performance coefficient of the chiller; G cw represents the chilled water flow rate passing through the cold coil, kg / s; G cw,min and G cw,max represent the minimum and maximum values of the chilled water flow rate passing through the cold coil, kg / s; β i (i = 0, 1, …, 15), γ i (i = 0, 1, …, 15) respectively represent the fitting coefficients of the temperature and humidity of the air after being treated by the cold coil. These coefficients can be obtained through simulation software or experimental data;

[0188] In the said objective function, the equivalent energy consumption of the hot coil is expressed by the following formula, that is, by adjusting the opening of the hot water valve and the hot water flow rate, the temperature of the air after being treated by the hot coil is changed to adjust the equivalent energy consumption of the hot coil:

[0189]

[0190] s.t. T sa = f coil (T ca , W ca , G sa , G hw )

[0191] = T ca + G hw [β1T ca + β2W ca + β3G sa + β4G hw + β5 + β6T ca2 + (48)

[0192] β7W ca 2 ++β8G sa 2 +β9G hw 2 +β 10 T ca W ca +β 11 W ca G sa +

[0193] β 12 G sa G hw +β 13 G hw T ca +β 14 T ca G sa +β 15 W ca G hw

[0194]

[0195] In the formula: the subscript hc represents the heat coil, and hw represents the hot water passing through the heat coil; η hc represents the efficiency of the heat coil; COP h represents the coefficient of performance of the boiler; G hw represents the hot water flow rate passing through the heat coil, kg / s; G hw,min and G hw,max represent the maximum value of the hot water flow rate passing through the heat coil, kg / s;

[0196] The objective function (5) can be normalized to obtain the following formula:

[0197]

[0198] In the formula: F min and F max are the minimum and maximum values of the objective function respectively;

[0199] In the model predictive control, the reference trajectory is obtained by the following formula:

[0200]

[0201] In the formula: represents the set values of the indoor temperature and moisture content; α is the softening coefficient, 0 < α < 1.

[0202] Such as Figure 5 ​As shown, preferably, adjusting the control parameters and minimizing the objective function through the intelligent optimization algorithm based on the national evolution process, and outputting the optimal control parameters, including:

[0203] S1: Initialize the country

[0204] Determine the policy parameters for the initially established country and set the number of policy parameters as D. Use Equation (21) as the objective function to calculate the fitness value of the initial policy parameters; The fitness value of the initial policy parameters is calculated using Equation (21).

[0205] The parameters of the initial policy can be expressed as:

[0206]

[0207] In the formula: x d represents the value of the d-th dimension in the policy parameters;

[0208] S2: The country determines the preliminary policy adjustment

[0209] Adjust the parameters by a large or small margin. Generate a random number R uniformly distributed between (0, 1). If R > 0.8, make a large adjustment; otherwise, make a small adjustment. After adjusting the policy parameters, calculate the fitness value of the new parameters;

[0210] Large adjustment: Use Levy flight to randomly generate new parameters. The method is as follows:

[0211]

[0212]

[0213] In the formula: u ∼ N(0, σ 2 ), v ∼ N(0, 1), and usually β = 1.5;

[0214] The new parameters are calculated as follows:

[0215]

[0216] If x' d exceeds the parameter range, then:

[0217] x'' d = |x' d - x' d,int |·(x d,max - x d,min ) + x d,min (56)

[0218] In the formula: x'' d is the d-th new parameter, and x' d,int is x' dThe integer part of, x d,max and x d,min are respectively the upper limit and the lower limit of the range of the d-th dimensional data;

[0219] Minor adjustment: In this embodiment, taking the parameters of the corresponding dimension of the large country as a reference, a random number that satisfies the normal distribution of x~N(0,1) is added, and this is used as the new parameter, that is:

[0220] x′ d = x d + x(57)

[0221] In the formula: x′ d is the new parameter of the d-th dimension; x d is the original parameter of the d-th dimension; x~N(0,1);

[0222] If x′ d exceeds the parameter range, then there is:

[0223]

[0224] In the formula, x″ is the new parameter of the d-th dimension, x d,max and x d,min are respectively the upper limit and the lower limit of the range of the d-th dimensional data;

[0225] S3: Determine whether the fitness of the policy parameters after the preliminary adjustment has improved

[0226] If the fitness of the parameters has improved after the preliminary adjustment, execute step S2 and continue with the policy adjustment; otherwise, execute step S4;

[0227] S4: The country updates the policy parameters

[0228] The country takes the parameter with the best fitness in the preliminary policy adjustment as the new national policy;

[0229] S5: The country splits into small countries

[0230] The large country splits into n small countries. Each small country randomly retains some of the parameters of the large country, and the remaining parameters are slightly adjusted in this embodiment in the same way as the minor adjustment method in S2. The policy parameters established by the small countries can be expressed as:

[0231] X i =(x i,1 ,…,x i,d ,…,x i,D )(59)

[0232] s.t.d∈[1,D],i∈[1,n]

[0233] In the formula: x i,d represents the value of the policy parameter of the d-th dimension of the i-th small country;

[0234] If the fitness of the newly generated parameters becomes worse compared to the fitness of the large country's parameters, the parameters of the small country will not be adjusted, and all parameters of the large country will be retained.

[0235] S6: Each small country sets up official positions and divides responsibilities

[0236] Each small country sets up m officials, among which 40% are ordinary court officials, 40% are spies, and 20% are prime ministers;

[0237] Define responsibilities for each official: Ordinary court officials put forward suggestions and make policy adjustments. Spies randomly go to other small countries to learn better policies from the court officials or prime ministers of other countries. The prime minister widely listens to the suggestions of the court officials and spies in his own country, learns from the better policies among the court officials and spies, and finally obtains the best policy parameters as the new policy parameters of the small country;

[0238] Each official has to adjust the policy parameters of each small country, that is, the initial parameters of each official are as shown in formula (29);

[0239] S7: Court officials put forward suggestions and make policy adjustments

[0240] Each court official in each small country starts to put forward suggestions and self-adjust the policies of each small country. Each court official generates a random number R uniformly distributed between (0, 1). If R > 0.8, a large adjustment will be made, otherwise a small adjustment will be made, and calculate the fitness value of the parameters after each adjustment;

[0241] The large adjustment method uses Levy flight to generate new parameters, and the small adjustment method uses normal distribution to generate new random numbers. The calculation steps are the same as S2. Each court official makes c policy adjustments and records the parameters with the best fitness in the policy adjustments of each court official;

[0242] S8: Spies go to other countries to learn for the first time

[0243] Each spy in each small country randomly selects another small country and learns the policy parameters put forward by the court officials of other countries in a certain way, and calculates the fitness value of the parameters after learning;

[0244] Use the PSO particle swarm optimization algorithm for learning. The spy selects the best parameters among the court officials of other countries as the learning object. That is, when initializing the particles in the PSO algorithm, each particle has an 80% probability of taking the learning object parameters as the initial position of the particle and a 20% probability of randomly initializing the position, and then perform PSO iterative calculation. Output the optimized best particle position as the new policy parameters that the spy can adopt after learning, and compare the fitness of this parameter with the fitness of the spy's original parameters. If it becomes worse, the spy does not adopt the suggestions of the court officials of other countries, and the policy parameters remain unchanged;

[0245] The above process is repeated s times, that is, the spies randomly select s small countries for policy learning and record the set of best parameters after the first learning of the spies in each small country;

[0246] S9: The prime minister initially collects the opinions of all officials

[0247] Each prime minister in each small country selects the top 30% of the best policy parameters among the court officials and spies in their respective small countries as the learning objects and learns from them and adjusts the initial policy parameters of the small country in a certain way;

[0248] The same PSO particle swarm optimization algorithm as in step S8 is used for learning. Each prime minister randomly selects a certain parameter among the top 30% of the best policy parameters of the court officials and spies in their own country as the learning object, performs PSO iterative calculation, and outputs the optimized best particle position as the new policy parameter after the prime minister's learning. This process is repeated p times, and the best parameters obtained during the p learning processes are used as the new parameters of the prime minister. Record the set of best parameters for the initial learning of the prime ministers in each small country.

[0249] S10: The spies go to other countries to learn for the second time

[0250] Each spy in each small country randomly selects another small country and learns from the policy parameters of the prime minister of the other country in a certain way, and calculates the fitness value of the learned parameters;

[0251] The PSO particle swarm optimization algorithm is used for learning. The specific steps are the same as in S8. The difference is that the spy randomly selects the best parameter among the prime ministers of other countries as the learning object. Record the set of best parameters for the second learning of the spies in each small country;

[0252] S11: The prime minister collects the opinions of all officials for the second time

[0253] Each prime minister selects the set of the top 30% of the best policy parameters among the court officials and spies in their respective small countries and learns from them and adjusts the policy parameters in a certain way;

[0254] The PSO particle swarm optimization algorithm is used for learning. The specific steps are the same as in S9. Record the set of best parameters for the second learning of the prime ministers in each small country;

[0255] S12: Obtain the best policy parameters among all small countries

[0256] Each small country updates its policy parameters: By comparing the fitness values of the parameters through the sets of best parameters obtained from the initial and second learning of the prime ministers in each small country, the best policy parameters among the prime ministers in each small country are obtained and used as the new policy parameters of each small country;

[0257] Compare the new policy parameters of each small country to obtain the parameter with the best fitness among all small countries, which can be expressed as:

[0258] X best =(x best,1,…,x best,d ,…,x best,D ) (60)

[0259] s.t. d ∈ [1, D]

[0260] where: x best,d represents the value of the d-th dimension of the best policy parameter among all small countries;

[0261] S13: The small country with the best policy parameter unifies all countries

[0262] The small country with the best policy parameter unifies all countries and becomes a new big country, taking its policy parameter as the initial policy parameter of the new country;

[0263] S14: Determine whether the iteration termination condition is reached

[0264] The iteration termination condition of this algorithm is to reach the maximum number of iterations or reach the specified convergence accuracy. If the iteration termination condition is reached, step S15 is executed; otherwise, S2 is executed.

[0265] S15: Output the best policy parameter

[0266] After continuous evolution of the country, finally the big country with the best policy parameter is output as the optimal value to complete the optimization calculation.

[0267] To verify the superiority of the optimal control method for a constant temperature and humidity system of the intelligent optimization algorithm based on the country evolution process (CEA) proposed in the present invention, 10 test functions are used for testing, and it is compared and verified with the particle swarm optimization algorithm (PSO), the satin bowerbird algorithm (SBO), and the sparrow search algorithm (SSA).

[0268] Under the AMD Ryzen 7 5800H with Radeon Graphics 3.20 GHz, 16.0 GB of memory, and the windows11 operating system, Pycharm2022.1.3 is used as the compiler to test the present invention.

[0269] The test functions are shown in Table 1, where functions F1 - F8 are unimodal functions and functions F9 - F12 are multimodal functions. The variable dimension of each test function is 30 - dimensional, and the maximum number of iterations is 300 times. Each algorithm runs independently on each test function 30 times, taking the average value as the optimization result and the standard deviation as the robustness evaluation index to reduce errors.

[0270] The population size of each algorithm is 30, and the specific parameters involved in PSO, SBO, SSA, and CEA - (the control method provided by the present invention) are shown in Table 2.

[0271] Each algorithm is separately used to perform individual operations on the test functions, and the mean and standard deviation of the calculation results are shown in Table 3.

[0272] Table 1 Test Functions

[0273]

[0274]

[0275] Table 2 Specific Parameters of the Algorithms

[0276]

[0277]

[0278] Table 3 Test Results of Each Algorithm

[0279]

[0280] As can be seen from Table 3, the CEA algorithm, a kind of intelligent optimization control method for predicting a constant temperature and humidity system based on the national evolution process proposed by the present invention, has obvious superiority compared with the other three algorithms.

[0281] The CEA algorithm is 4 to 49 orders of magnitude higher than the other three algorithms in the convergence accuracy of unimodal functions F1 - F8, all below 1e - 10; the optimization results of the CEA algorithm for multimodal functions F9 and F10 are the optimal values of the functions; the CEA algorithm is 5 to 24 orders of magnitude higher than the other three algorithms in the convergence accuracy of multimodal functions F11 and F12, and all below 1e - 11. It can be seen that the CEA algorithm has good optimization accuracy.

[0282] The order of magnitude of the standard deviation of the CEA algorithm is equal to or less than the order of magnitude of its optimization result, and at the same time is much higher than the order of magnitude of the standard deviation of other algorithms. Especially, the standard deviation during the optimization process of multimodal functions F9 and F10 is 0, indicating that the CEA algorithm has good robustness.

[0283] The convergence curves of each algorithm are as Figure 6 shown. As Figure 6 can be seen, the CEA algorithm has a very fast convergence speed. Except for F5 and F9, for other test functions, a high convergence accuracy can be achieved within 25 iterations. Especially during the iteration process of multimodal function F10, the global optimal value can be found within 5 iterations. The CEA algorithm can also achieve a high convergence accuracy within 100 iterations for unimodal function F5, and at the same time still has good search ability in the later stage of iteration. For multimodal function F9, the global optimal value can be found after about 50 iterations. Compared with the other three algorithms, the CEA algorithm can jump out of the local optimum at the initial stage of iteration, with a fast convergence speed, showing high optimization ability.

[0284] In summary, the CEA algorithm has achieved good results for the test functions listed in Table 2, demonstrating excellent global search ability and robustness. Through analysis, it can be seen that each court official randomly puts forward suggestions, avoiding the simplification of national policies and making it more likely to obtain better policies. That is, the random suggestions of court officials prevent the algorithm from falling into local optima; the spies randomly go to other small countries to learn better national policies proposed by court officials or prime ministers in other countries, enabling the spies to obtain the most likely optimal policy plan and improving the policy level of the country where the spies are located. That is, the learning behavior of the spies improves the global search ability of the algorithm; the prime minister listens to the relatively good policies obtained by court officials and spies in his own country and then thinks out the best policy. That is, the behavior of the prime minister adopting the admonitions of all officials to find better policies enables the algorithm to conduct fine optimization near the relatively good solutions.

[0285] In this embodiment, the model predictive control method combined with the CEA algorithm and the PI control method are respectively used to control the constant temperature and humidity of a certain room. The room temperature is set to 23 ± 2 °C, and the relative humidity is set to 60 ± 10%.

[0286] Appendix Figure 7 and Appendix Figure 8 show the temperature and relative humidity change trend graphs of the room under the two schemes. It can be seen that under the model predictive control combined with the CEA algorithm, the room temperature fluctuation range is within 2 °C, and the maximum deviation is 2 °C, meeting the indoor temperature setting requirements; the room relative humidity fluctuation range is within 10%, and the maximum deviation is 8%, meeting the indoor relative humidity setting requirements. Under the PI control strategy, however, the fluctuation ranges of indoor temperature and relative humidity are both relatively large. Among them, the maximum deviation of indoor temperature reaches 2.5 °C, and the maximum deviation of relative humidity reaches 20%, both of which do not meet the setting requirements.

[0287] Appendix Figure 9 shows the power change graph of the system over time under the two schemes. Under the PI control strategy, the system power fluctuates around 750 kW; under the model predictive control combined with the CEA algorithm, the system power fluctuates around 700 kW, reducing the system energy consumption.

[0288] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An optimized control method for a constant temperature and humidity system, characterized in that, Including: Collect and obtain data, input the collected data into a neural network model, and train the neural network system prediction model; The data collection and acquisition includes: outdoor air temperature and moisture content, supply air temperature and moisture content, indoor air temperature and moisture content, fan frequency, cold coil water valve opening, hot coil water valve opening, room cooling load, and room moisture load; The room cooling load and room moisture load in the acquired data can be calculated by the following formula, where the solar radiation irradiance is collected by a solar radiation sensor; Where: represents the room cooling load, in W; G sa represents the supply air volume, in kg / s; c a represents the specific heat capacity of air, in J / (kg·°C); T sa represents the supply air temperature, in °C; T Z represents the indoor air temperature at the current moment, in °C; k wall represents the heat transfer coefficient of the wall, in W / (m 2 ·°C); A wall represents the area of the wall, in m 2 ; T oa represents the outdoor air temperature, in °C; A e represents the effective area of the building, in m 2 ; I represents the solar radiation irradiance, in W / m 2 ; represents the cooling load formed by people in the room per unit time, in W; represents the cooling load formed by other factors in the room, such as lamps and equipment, in W; W represents the moisture content, in kg / (kg dryair); represents the relative humidity, in %; P q,b represents the saturated water vapor pressure, in Pa; B represents the atmospheric pressure, in Pa; represents the room moisture load, in kg / s; W sa represents the moisture content of the supply air, in kg / (kg dryair); W z represents the moisture content of the indoor air in kg / (kg dryair); represents the moisture load of people in the room, in kg / s; During the training process of the neural network, the mean square error (MSE) is used to evaluate the training effect: Wherein: represents the target output parameter; represents the actual output parameter; N represents the number of samples; Design a rolling optimizer, determine the prediction domain and control domain of the system prediction model controller, and set the objective function, including: The objective function consists of the tracking errors of the predicted indoor air temperature and indoor air moisture content, and the equipment energy consumption. Among them, the equipment energy consumption consists of fan energy consumption, chiller power consumption, and reheater power consumption; F3 = P fan +P cc +P hc (8) f min <f<f max (9) L cc,min <L cc <L cc,max (10) L hc,min <L hc <L hc,max (11) Where: J represents the objective function, N p represents the prediction domain; w i represents the weight, i = 1, 2, 3; F1 represents the tracking error of predicting the indoor air temperature; F2 represents the tracking error of predicting the indoor air moisture content; F3 represents the equipment energy consumption; represents the indoor air temperature predicted by the model, °C; T z,r (k + i) represents the reference trajectory of the indoor air temperature, °C; represents the indoor air moisture content predicted by the model, kg / (kg dry air); W z,r (k + i) represents the reference trajectory of the indoor air moisture content, kg / (kg dry air); the subscript fan represents the fan, cc represents the cooling coil, hc represents the heating coil, min represents the minimum value, max represents the maximum value; P represents the equipment energy consumption, W; f represents the fan frequency, Hz; L represents the opening of the water valve, %; For the constant temperature and humidity air conditioning system, a fresh air condition is adopted, and there are linear relationships between the supply air volume and the fan frequency, and between the water flow rate and the water valve opening; In the objective function, the equivalent energy consumption of the cold coil is expressed by the following formula, that is, by adjusting the opening of the cold water valve, adjusting the chilled water flow rate, thereby changing the enthalpy value of the air after being processed by the cold coil, and adjusting the equivalent energy consumption of the cold coil; h ca = 1.01T ca + W ca (2501 + 1.84T ca ) (15) In the formula: s.t. is a mathematical abbreviation, representing "subject to", constrained. The subscript cc represents the cold coil, sa represents the supply air, oa represents the outdoor air, ca represents the air after being processed by the cold coil, and cw represents the chilled water passing through the cold coil; h oa represents the enthalpy value of the outdoor air. When the operating condition is the return air condition, this item is the enthalpy value of the air after mixing fresh air and return air, J; h ca represents the enthalpy value of the air after being processed by the cold coil, J; T ca represents the temperature of the air after being processed by the cold coil, °C; W ca represents the moisture content of the air after being processed by the cold coil, kg / (kg dry air); W oa represents the moisture content of the outdoor air, kg / (kg dry air); η cc represents the efficiency of the cold coil; COP c represents the performance coefficient of the chiller; G cw represents the chilled water flow rate passing through the cold coil, kg / s; G cw,min and G cw,max represent the minimum and maximum values of the chilled water flow rate passing through the cold coil, kg / s; β i (i = 0, 1, …, 15), γ i (i = 0, 1, …, 15) respectively represent the fitting coefficients of the temperature and humidity of the air after being processed by the cold coil. These coefficients are obtained through simulation software or experimental data; In the objective function, the equivalent energy consumption of the hot coil is expressed by the following formula, that is, by adjusting the opening of the hot water valve, adjusting the hot water flow rate, thereby changing the temperature of the air after being processed by the hot coil, and adjusting the equivalent energy consumption of the hot coil: In the formula: the subscript hc represents the heat coil, and hw represents the hot water passing through the heat coil; η hc represents the efficiency of the heat coil; COP h represents the coefficient of performance of the boiler; G hw represents the hot water flow rate passing through the heat coil, kg / s; G hw,min and G hw,max represent the maximum value of the hot water flow rate passing through the heat coil, kg / s, and s.t. is a mathematical abbreviation representing subject to, subject to constraints; The objective function (5) after normalization is obtained as the following formula: Where: F 1,min represents the minimum value of the tracking error of the predicted indoor air temperature, and F 1,max, represents the maximum value of the tracking error of the predicted indoor air temperature; F 2,min represents the minimum value of the tracking error of the predicted indoor air moisture content, and F 2,max represents the maximum value of the tracking error of the predicted indoor air moisture content; F 3,min represents the minimum value of the equipment energy consumption, and F 3,max represents the maximum value of the equipment energy consumption; In the model predictive control, the reference trajectory is obtained by the following formula: In the formula: represents the set values of indoor temperature and moisture content; α is the softening coefficient, where 0 < α < 1; The input parameters of the prediction model include: the indoor air temperature T at the current moment z (k), the moisture content W of the indoor air z (k), the supply air temperature T sa (k), the moisture content W of the supply air sa (k), the room cooling load and the room moisture load the fan frequency f(k - 1) at the previous moment, the opening degree L of the cold coil water valve cc (k - 1), the opening degree L of the hot coil water valve hc (k - 1); The control module obtains the input parameters of the prediction model during the operation of the system through various sensors, inputs them into the neural network system prediction model, adjusts the control parameters through an intelligent optimization algorithm based on the national evolution process, minimizes the objective function, and outputs the optimal control parameters; The control module transmits the output optimal control parameters to the chiller water valve controller, reheater water valve controller, and fan frequency converter in the form of digital signals, so as to realize the adjustment of the fan frequency, chilled water flow rate, and hot water flow rate, and change the supply air volume, supply air temperature, and supply air humidity.

2. The optimized control method for a constant temperature and humidity system according to claim 1, characterized in that The adjustment of the control parameters through the intelligent optimization algorithm based on the national evolution process and the minimization of the objective function to output the optimal control parameters includes: S1: Initialize the country Determine policy parameters for the initially established country and set the number of policy parameters as D. Use Equation (21) as the objective function to calculate the fitness value of the initial policy parameters; ​ The parameters of the initial policy can be expressed as: s.t.d∈[1,D] where: x d represents the value of the d-th dimension in the policy parameters; S2: Determine the preliminary policy adjustment of the country Adopt large or small adjustments to the parameters, generate a random number R uniformly distributed between (0,1). If R>0.8, make a large adjustment; otherwise, make a small adjustment. After adjusting the policy parameters, calculate the fitness value of the new parameters; Large adjustment: Use Levy flight to randomly generate new parameters, and the method is as follows: where: u ~ N(0, σ 2 ), v ~ N(0, 1), and usually β = 1.5; The new parameters are calculated as follows: If x' d exceeds the parameter range, then there is: where: x″ d is the new parameter of the d-th dimension, x′ d,int is the integer part of x′ d x d,max and x d,min are the upper limit and the lower limit of the range of the d-th dimensional data respectively; Small adjustment: In this embodiment, the parameters corresponding to the large country in the corresponding dimension are used as the benchmark, and a random number satisfying x~N(0,1) is added as the new parameter, that is: x′ d = x d + x(27) where: x' d is the new parameter of the d-th dimension; x d is the original parameter of the d-th dimension; x ∼ N(0, 1); If x' d exceeds the parameter range, then there is: where x″ is the new parameter of the d-th dimension, and x d,max and x d,min are the upper limit and the lower limit of the range of the d-th dimension data, respectively; S3: Judge whether the fitness of the policy parameters after the preliminary adjustment becomes better If the fitness of the parameters improves after the preliminary adjustment, step S2 is executed to continue the policy adjustment; otherwise, step S4 is executed; S4: The country updates the policy parameters The country takes the parameters with the best fitness in the preliminary policy adjustment as the new national policy; S5: The country splits into small countries The large country splits into n small countries. Each small country randomly retains some of the parameters of the large country, and the remaining parameters are slightly adjusted in this embodiment in the same way as the small adjustment method in S2. The policy parameters established by the small countries can be expressed as: X i =(x i,1 ,…,x i,d ,…,x i,D )(29) s.t.d∈[1,D],i∈[1,n] where: x i,d represents the policy parameter value of the i-th small country in the d-th dimension; If the fitness of the newly generated parameters becomes worse compared with the fitness of the large country, the parameters of the small countries are not adjusted, and all the parameters of the large country are retained; S6: Each small country assigns official positions Each small country establishes m officials, among which 40% are ordinary court officials, 40% are spies, and 20% are prime ministers; Define responsibilities for each official: Ordinary court officials put forward suggestions and conduct policy adjustments. Spies randomly go to other small countries to learn better policies from the court officials or prime ministers of other countries. The prime minister widely listens to the suggestions of the court officials and spies in his own country, learns from the better policies among the court officials and spies, and finally obtains the best policy parameters as the new policy parameters of the small country; Each official has to adjust the policy parameters of each small country, that is, the initial parameters of each official are as shown in formula (29); S7: Court officials put forward suggestions for policy adjustment Each court official in each small country starts to put forward suggestions and self-adjusts the policies of each small country. Each court official generates a random number R uniformly distributed between (0,1). If R>0.8, a large adjustment is made; otherwise, a small adjustment is made, and the fitness value of the parameters after each adjustment is calculated; The large adjustment method uses Levy flight to generate new parameters, and the small adjustment method uses the normal distribution to generate new random numbers. The calculation steps are the same as in S2. Each court official makes c policy adjustments and records the parameters with the best fitness in the policy adjustments of each court official; S8: Spies go to other countries to learn for the first time Each spy in each small country randomly selects another small country and learns the policy parameters proposed by the court officials of other countries in a certain way, and calculates the fitness value of the parameters after learning; The PSO particle swarm algorithm is used for learning. The spy selects the best parameters among the court officials of other countries as the learning object. That is, when initializing the particles in the PSO algorithm, each particle has an 80% probability of taking the learning object parameters as the initial position of the particle and a 20% probability of randomly initializing the position, and then performs PSO iterative calculation; the optimized best particle position is output as the new policy parameters that can be adopted after the spy learns. Compare the fitness of this parameter with the fitness of the spy's original parameters. If it becomes worse, the spy does not adopt the suggestions of the court officials of other countries, and the policy parameters remain unchanged; The above process is repeated s times, that is, the spy randomly selects s small countries for policy learning, and records the set of the best parameters after the first learning of the spies in each small country; S9: The prime minister collects the opinions of all officials for the first time Each prime minister in each small country selects the top 30% of the best policy parameters among the court officials and spies in the small country where he is located as the learning object, and learns and adjusts the initial policy parameters of the small country in a certain way; Use the same PSO particle swarm optimization algorithm as in step S8 for learning. Each prime minister randomly selects a certain parameter from the top 30% of the best policy parameters among the courtiers and spies in his own country as the learning object, performs PSO iterative calculation, and outputs the optimized best particle position as the new policy parameter after the prime minister's learning. This process is repeated p times, and the best parameters obtained during the p learning processes are used as the new parameters of the prime minister; record the set of best parameters for the initial learning of the prime ministers of each small country; S10: The spies go to other countries to learn for the second time Each spy in each small country randomly selects another small country and learns from the policy parameters of the prime minister of the other country in a certain way, and calculates the fitness value of the parameters after learning; Use the PSO particle swarm optimization algorithm for learning. The specific steps are the same as in S8. The difference is that the spies randomly select the best parameters among the prime ministers of other countries as the learning object; record the set of best parameters for the second learning of the spies in each small country; S11: The prime ministers collect the opinions of all officials for the second time Each prime minister selects the set of the top 30% of the best policy parameters among the courtiers and spies in the small country where he is located, and learns from and adjusts the policy parameters in a certain way; Use the PSO particle swarm optimization algorithm for learning. The specific steps are the same as in S9, and record the set of best parameters for the second learning of the prime ministers of each small country; S12: Obtain the best policy parameters among all small countries Each small country updates its policy parameters: through the sets of best parameters for the initial and second learning of the prime ministers of each small country, compare the fitness values of the parameters to obtain the best policy parameters among the prime ministers of each small country, and use them as the new policy parameters of each small country; Compare the new policy parameters of each small country to obtain the parameter with the best fitness among all small countries, which can be expressed as: X best = (x best,1 , …, x best,d , …, x best,D ) (30) s.t.d∈[1,D] where: x best,d represents the value of the d-th dimension of the optimal policy parameter among all small countries; s.t. is a mathematical abbreviation meaning subject to S13: The small country with the best policy parameters unifies all countries Having the optimal policy parameters A small country with these unifies all countries and becomes a new large country, taking its policy parameters as the initial policy parameters of the new country; S14: Determine whether the iteration termination condition is reached The iteration termination condition of this algorithm is to reach the maximum number of iterations or reach the specified convergence accuracy. If the iteration termination condition is reached, execute step S15; otherwise, execute S2; S15: Output the best policy parameters After continuous evolution of the country, the major country with the best policy parameters is output as the optimal value to complete the optimization calculation.

Images