Incubator with multiple independent temperature-controlled chambers and artificial intelligence temperature control system

Abstract

The application relates to a culture box with multiple independent temperature-controllable box bodies and an artificial intelligence temperature control system, and belongs to the technical field of automatic control. The artificial intelligence temperature control system comprises N artificial intelligence subsystems for controlling the temperatures of N box bodies respectively. In each artificial intelligence subsystem, a subtracter is used for subtracting a set temperature from a measured temperature in a box body i to obtain an error signal, a reinforcement learning module i generates a state of the box body i according to the error signal, generates a control strategy of a PID controller i of an electronic expansion valve i or a heating element i at time t according to the state, and the reinforcement learning module also generates a control strategy of an electronic switch i at time t according to the error signal. The application can control the temperatures of multiple box bodies by using one compressor, so that the cost is reduced, and the efficiency is improved.

Description

Technical Field

[0001] This invention relates to an incubator with multiple independent temperature control chambers and an artificial intelligence temperature control system, belonging to the field of automatic control technology. Background Technology

[0002] Traditional split-type air conditioning systems rely on a single compressor for centralized cooling, which cannot independently adjust the cooling capacity of each zone, resulting in large temperature fluctuations. Dual-unit cascade compressors are expensive and complex to maintain, and have low energy efficiency ratios at low temperatures; their refrigerant flow control accuracy is insufficient, and their thermostatic expansion valves have slow response times, making them unable to adapt to dynamic load changes. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, this invention provides an incubator with multiple independent temperature-controlled chambers and an artificial intelligence temperature control system. It can control the temperature of multiple chambers with a single compressor, thereby reducing costs and improving efficiency.

[0004] To achieve the aforementioned objective, this invention provides an incubator with multiple independently temperature-controlled chambers, comprising: a refrigerant supply module, N chambers, and an artificial intelligence temperature control system. The i-th chamber is equipped with a temperature sensor i, an evaporator i, a heating wire i, and an electronic expansion valve i. The refrigerant supply module is connected to the evaporator i via the electronic expansion valve i. The heating wire i provides heat energy to the chamber i. The temperature sensor i is configured to acquire a sequence of temperatures within the chamber i. The artificial intelligence temperature control system includes N artificial intelligence subsystems that control the temperatures of the N chambers respectively. The i-th artificial intelligence subsystem includes a reinforcement learning module i, a PID controller i, an electronic transfer switch i, and a subtractor i. The subtractor i is used to subtract the set temperature within the chamber i from the measured temperature to obtain an error signal. The reinforcement learning module i is based on the error signal Generate the state of box i According to the status The control strategy of generating the PID controller i that controls the electronic expansion valve i or the heating wire i at time t. ,in, Let i be the control parameter vector of the PID controller at time t. These represent the proportional, integral, and derivative coefficients of the PID controller i at time t; the reinforcement learning module i also considers the error signal... Control strategy for generating electronic switching switch i at time t , i=1,…,N, This represents the control signal of electronic switch i at time t.

[0005] To achieve the aforementioned objective, this invention also provides an artificial intelligence temperature control system, comprising N artificial intelligence subsystems that respectively control the temperature of N chambers. The i-th artificial intelligence subsystem includes a reinforcement learning module i, a PID controller i, an electronic transfer switch i, and a subtractor i. The subtractor i is used to subtract the set temperature inside chamber i from the measured temperature to obtain an error signal. The reinforcement learning module i is based on the error signal Generate the state of box i According to the status The control strategy of generating the PID controller i that controls the electronic expansion valve i or the heating wire i at time t. ,in, Let i be the control parameter vector of the PID controller at time t. These represent the proportional, integral, and derivative coefficients of the PID controller i at time t; the reinforcement learning module i also considers the error signal... Control strategy for generating electronic switching switch i at time t , i=1,…,N, This represents the control signal of electronic switch i at time t.

[0006] Compared with existing technologies, the incubator and artificial intelligence temperature control system provided by this invention have multiple independent temperature control chambers. The reinforcement learning module of the artificial intelligence subsystem of each chamber generates a state based on the error signal, and generates a control strategy for the PID controller that controls the electronic expansion valve or heating wire based on the state. The control strategy of the electronic switching switch is generated based on the error signal to select the electronic expansion valve or heating wire to be connected to the PID controller. In this way, the temperature of multiple chambers can be controlled by a single compressor, thereby reducing costs and improving efficiency. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of an incubator with multiple independent temperature-controlled chambers provided in the first embodiment of the present invention.

[0008] Figure 2 This is a block diagram of the artificial intelligence subsystem provided in the first embodiment of the present invention.

[0009] Figure 3 This is a schematic diagram of the neural network of the reinforcement learning module constructed in the first embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] First Embodiment

[0012] Figure 1 This is a schematic diagram of an incubator with multiple independent temperature-controlled chambers provided in the first embodiment of the present invention, as shown below. Figure 1 As shown, the incubator with multiple independent temperature-controlled chambers provided by the present invention includes a shell, within which N independent chambers are included. An evaporator 6 is disposed on the outer side of one side wall of each chamber, and a heating wire (not shown) is disposed on the outer side of the other side wall or the outer side of the bottom wall. The incubator also includes N temperature sensors, with the probe of each temperature sensor extending into one chamber. The incubator further includes a refrigerant supply module and N electronic expansion valves, such as electronic expansion valve 7, electronic expansion valve 8, and electronic expansion valve 9. The refrigerant supply module includes a fixed-frequency compressor 1, a condenser 2, and a capillary tube 5. The fixed-frequency compressor 1 draws in low-temperature, low-pressure refrigerant gas and compresses it into a high-temperature, high-pressure gaseous refrigerant. The condenser receives the high-temperature, high-pressure gaseous refrigerant and, through a copper tube fin structure and a fan, forces heat dissipation, condensing the refrigerant into a high-pressure liquid and releasing heat to the outside. The capillary tube 5 throttles and reduces the pressure of the high-pressure liquid refrigerant, converting it into a low-temperature, low-pressure refrigerant, which is then delivered to N evaporators through N electronic expansion valves. The low-temperature, low-pressure refrigerant absorbs heat from its corresponding chamber and evaporates into a gaseous state within the evaporator, simultaneously lowering the air temperature and dehumidifying the air. The evaporators are designed as coils. Optionally, a drying tube 4 is installed between the condenser and the capillary tube, containing a desiccant (such as molecular sieves or silica gel) to absorb moisture from the refrigerant. Moisture can cause freezing and blockage of the capillary tube and electronic expansion valves at low temperatures, leading to system malfunctions; simultaneously, the acidic substances produced by the reaction of moisture and refrigerant can corrode the pipes. The drying tube is equipped with a filter screen that intercepts impurities such as metal shavings, installation residue, or compressor wear powder. If impurities enter the capillary tube or the fixed-frequency compressor, they can cause blockages or mechanical damage; the drying tube's filtration function prevents such problems. The drying tube also has liquid storage capacity. The evaporator is connected to the electronic expansion valve via a high-pressure line and to the fixed-frequency compressor via a low-pressure gas line 3.

[0013] Each chamber is also equipped with a fan, such as fan 10, which is used to regulate the airflow inside the chamber and further regulate the temperature in various parts of the chamber.

[0014] In the first embodiment, the incubator also includes an artificial intelligence temperature control system, which comprises N artificial intelligence subsystems that respectively control the temperature of N chambers. The following describes the subsystems in conjunction with... Figure 2 The composition and working process of each artificial intelligence subsystem will be explained in detail.

[0015] Figure 2 This is a block diagram of the artificial intelligence subsystem provided in the first embodiment of the present invention, as shown below. Figure 2 As shown, the i-th artificial intelligence subsystem includes a reinforcement learning module i, a PID controller i, an electronic transfer switch i, and a subtractor i. The subtractor i is used to subtract the set temperature inside the chamber i from the measured temperature to obtain an error signal. The reinforcement learning module i is based on the error signal Generate the state of box i According to the status The control strategy of generating the PID controller i that controls the electronic expansion valve i or the heating wire i at time t. ,in, Let i be the control parameter vector of the PID controller at time t. These represent the proportional, integral, and derivative coefficients of the PID controller i at time t; the reinforcement learning module i also considers the error signal... Control strategy for generating electronic switching switch i at time t , i=1,…,N, This represents the control signal of electronic switch i at time t.

[0016] In the first embodiment, the electronic expansion valve i and the heating wire i are connected to the output terminal through the adder i. The adder i can also be replaced by a connection node.

[0017] In the first embodiment, ,

[0018] In the formula, These represent the proportional, integral, and derivative coefficients of the PID controller i at time t; the electronic expansion valve or heating wire is based on the control parameter vector. Different degrees of opening and closing or power selection can be performed. In the first embodiment, the electronic expansion valve and the heating wire do not work simultaneously. When cooling is required, the electronic expansion valve works; when heating is required, the heating wire works.

[0019] Still Figure 2 As shown, the output of PID controller i at time t is The set temperature inside the enclosure at time t. and the measured temperature at time t The error is :

[0020] ,

[0021] In the formula, , ; ; For box i in time The set temperature at that time and in time Actual measured temperature at time Error; For box i in time The set temperature at that time and in time Actual measured temperature at time Error; The time step; in the first embodiment, the set temperature of chamber i is the same for the set time period, that is... In the formula, y id The set temperature for the set time period.

[0022] Written in matrix form:

[0023] ,

[0024] In the formula, , ,

[0025] In the formula, T represents transpose.

[0026] Figure 3 This is a schematic diagram of the neural network of the reinforcement learning module constructed in the first embodiment of the present invention, as shown below. Figure 3 As shown, the neural network includes an input layer, a hidden layer, and an output layer. The input layer consists of three neurons, each receiving an input vector. The three elements in the process; the hidden layer includes 3 neurons; the output layer includes 4 neurons, of which the first to third neurons output the proportional, integral and derivative coefficients of the PID controller, respectively, and the fourth neuron outputs the state value function.

[0027] In the first embodiment, the reinforcement learning module i determines the state of the box i. Generate the control strategy for PID controller i at time t include:

[0028] S1-1: Calculate the coefficient at time t according to the following formula. :

[0029] ,

[0030] In the formula, Let j be the output function of the j-th neuron in the hidden layer at time t. and is the center and width of the jth Gaussian function of the neural network of reinforcement learning module i at time t, the setting of the center determines how quickly the neural network adapts to the change of , directly affects the speed of ; is the weight between the jth neuron of the hidden layer and the nth neuron of the output layer of the neural network of reinforcement learning module i at time t, which is used to adjust the connection strength between neurons to affect the speed of information flow, while the weight affects the computing resources, directly affecting the control consumption; in addition, the contribution of the weight to the total output signal affects the signal level; j = 1, 2, 3; n = 1, 2, 3, represents the two-norm.

[0031] The reinforcement learning module i generates the control strategy of the PID controller i at time t according to the state of the box i , which includes:

[0032] S1-2: Calculate the state value function at time t according to the following formula :

[0033] ,

[0034] In the formula, is the weight between the jth neuron of the hidden layer and the Nth neuron of the output layer of the neural network of reinforcement learning module i at time t, which is used to adjust the connection strength between neurons to affect the speed of information flow, while the weight affects the computing resources, directly affecting the control consumption; in addition, the contribution of the weight to the total output signal affects the signal level, N = 4.

[0035] The reinforcement learning module i generates the control strategy of the PID controller i at time t according to the state of the box i , which also includes:

[0036] S1-3: Build a loss function according to the following formula :

[0037] ,

[0038] In the formula, is the set temperature of the box i at time t, is the measured temperature of the box i at time t.

[0039] The reinforcement learning module i generates the control strategy of the PID controller i at time t according to the state of the box i , which also includes:

[0040] ​​​S1-4: update parameters according to the following equations:

[0041] ,

[0042] ,

[0043] ,

[0044] ,

[0045] wherein, is the reward obtained by the PID controller i; , , , is the learning coefficient; is the weight between the jth neuron in the hidden layer and the nth neuron in the output layer of the neural network of the reinforcement learning module i at time t, n = 1, 2, 3; is the weight between the jth neuron in the hidden layer and the Nth neuron in the output layer of the neural network of the reinforcement learning module i at time t, N = 4; is the center of the jth Gaussian function of the neural network of the reinforcement learning module i at time t; is the bandwidth of the jth Gaussian function of the neural network of the reinforcement learning module i at time t; denotes the gradient of denotes the gradient of denotes the gradient of denotes the gradient of denotes the gradient of denotes the gradient of is the discount coefficient; is the time step.

[0046] In the first embodiment,

[0047] ;

[0048] ;

[0049] ;

[0050] ;

[0051] .

[0052] ​​​​​​​Reinforcement learning module i depends on the state of box i Generate the control strategy for PID controller i at time t Also includes:

[0053] S1-5: Determine the loss function Is it the smallest? If not, , , , Then return to step S1-1; if yes, output , , , As the optimal parameter for calculation , , .

[0054] In the first embodiment, the reinforcement learning module i also bases its learning on the error signal. Control strategy for generating electronic switching switch i at time t Includes: judgment The value is either positive or negative. If negative, the PID controller i is connected to the electronic expansion valve i via the electronic switch i. The electronic expansion valve i operates according to the control of the PID controller i. The refrigerant supply module supplies low-temperature, low-pressure refrigerant to the evaporator located on the outer side wall of the housing i through the electronic expansion valve i. The low-temperature, low-pressure refrigerant absorbs heat from the corresponding area inside the housing i in the evaporator and evaporates into a gaseous state, thus lowering the temperature inside the housing i. If positive, the PID controller i is connected to the heating wire i via the electronic switch K. The heating wire i operates according to the control of the PID controller i, heating the housing i and raising the temperature inside the housing i. If... If this happens, neither the electronic expansion valve i nor the heating wire i will work.

[0055] Compared with the prior art, the split-type independent temperature control device and artificial intelligence temperature control system of the incubator provided by the present invention in the first embodiment have the following beneficial effects: the reinforcement learning module of the artificial intelligence control subsystem of each chamber generates a state based on the error signal, and generates a control strategy based on the state to control the electronic expansion valve of the cooling evaporator of the chamber or the heating wire of the heating element for heating the chamber; the control strategy of the electronic switching switch is generated based on the error signal to select the electronic expansion valve or the heating element to be connected to the PID controller, thereby achieving the goal of controlling the temperature of multiple chambers with one compressor, thereby reducing costs and improving efficiency.

[0056] Furthermore, although the parameters of the neural network in the reinforcement learning module are dynamically updated over time using gradient descent, the initial selection of these parameters is crucial for achieving the desired results.

[0057] Second embodiment

[0058] The second embodiment of the present application only describes the different content from the first embodiment, and the same content is not repeated.

[0059] In the second embodiment, the reinforcement learning module i generates a control policy of the PID controller i for controlling the electronic expansion valve i or the heating i at time t according to the error signal The state of the box i According to the state The control policy of the PID controller i for controlling the electronic expansion valve i or the heating i at time t is generated Including:

[0060] S2-1: Calculate the state-action function at time t according to the following formula:

[0061] ,

[0062] In the formula, The transition probability of the state from the state to the state is adopted by the PID controller i at time t using the control parameter vector The state value function at time t is Indicates the discount factor; The reward obtained by the PID controller i using the control parameter vector ;

[0063] S2-2: Improve According to the following formula:

[0064] ,

[0065] In the formula, The time step is

[0066] S2-3: Calculate the state value function at time According to the following formula:

[0067] ;

[0068] S2-4: Determine Whether is less than or equal to the threshold value , if yes, output the control parameter vector of the PID controller i at time ; Otherwise, go back to step S2-1.

[0069] Compared with the prior art, the artificial intelligence temperature control system provided by the second embodiment has the beneficial effects that: the reinforcement learning module of the artificial intelligence control subsystem of each cabinet generates a control strategy of a PID controller of an electronic expansion valve of a cooling evaporator of the cabinet or a heating wire for warming up the cabinet according to a state generated according to an error signal; and the control strategy of an electronic switch is generated according to the error signal to select the electronic expansion valve or the heating wire to be connected to the PID controller, so that the temperature of multiple cabinets can be controlled by one compressor, thereby reducing the cost and improving the efficiency.

[0070] The disclosed preferred embodiments of the present application are only used to help explain the present application, the preferred embodiments do not describe all the details, and the present application is not limited to the specific embodiments. Obviously, many modifications and variations can be made according to the content of the present application. The present application is selected and specifically described, in order to better explain the principles and practical applications of the present application, so that those skilled in the art can well understand and utilize the present application. The present application is limited by the claims and their full scope and equivalents.

Claims

1. An artificial intelligence temperature control system for an incubator having a plurality of independently temperature controlled chambers, wherein, The incubator includes a refrigerant supply module and N chambers. The i-th chamber is equipped with a temperature sensor i, an evaporator i, a heating wire i, and an electronic expansion valve i. The refrigerant supply module is connected to the evaporator i via the electronic expansion valve i. The heating wire i provides heat energy to the chamber i, and the temperature sensor i is configured to acquire the sequential temperature within the chamber i. The artificial intelligence temperature control system includes N artificial intelligence subsystems that control the temperature of the N chambers respectively. The i-th artificial intelligence subsystem includes a reinforcement learning module i, a PID controller i, an electronic transfer switch i, and a subtractor i. The subtractor i is used to subtract the set temperature and the measured temperature within the chamber i to obtain an error signal. The feature is that the reinforcement learning module i, based on the error signal Generate the state of box i According to the status The control strategy of generating the PID controller i that controls the electronic expansion valve i or the heating wire i at time t. ,in, In the formula, Let i be the control parameter vector of the PID controller at time t. These represent the proportional, integral, and derivative coefficients of the PID controller i at time t; the reinforcement learning module i also considers the error signal... Control strategy for generating electronic switching switch i at time t , This represents the control signal of electronic switch i at time t, where i = 1, ..., N; This represents the control signal of electronic switch i at time t. wherein , wherein , and are the center and width of the jth Gaussian function of the neural network of the reinforcement learning module i at time t; are the weights between the jth neuron of the hidden layer of the neural network of the reinforcement learning module and the nth neuron of the output layer at time t; j = 1, 2, 3; n = 1, 2, 3, denotes the two-norm.

2. The artificial intelligence temperature control system of incubator with multiple independently temperature-controlled chambers of claim 1, wherein, The reinforcement learning module i computes the state value function at time t according to : ， wherein is the weight between the jth neuron of the hidden layer of the neural network of the reinforcement learning module i and the Nth neuron of the output layer at time t, N = 4.

3. The artificial intelligence temperature control system of incubator with multiple independently temperature-controlled chambers of claim 2, wherein, The reinforcement learning module i builds a loss function according to the following formula : ， wherein Tset is the set temperature within the cabinet i, Tmeas is the measured temperature within the cabinet i; Update the parameters according to the following formula: ， ， , ， wherein is the control parameter vector of the PID controller i and the reward obtained is , , , is the learning coefficient is the weight of the neural network of the reinforcement learning module i between the jth neuron of the hidden layer and the nth neuron of the output layer at time t, n = 1, 2, 3 is the weight of the neural network of the reinforcement learning module i between the jth neuron of the hidden layer and the Nth neuron of the output layer at time t, N = 4 is the center of the jth Gaussian function of the neural network of the reinforcement learning module i at time t is the bandwidth of the jth Gaussian function of the neural network of the reinforcement learning module i at time t denotes the gradient of denotes the gradient of denotes the gradient of denotes the gradient of is the discount factor is the time step​​​​​​​​ 4. An artificial intelligence temperature control system for an incubator having a plurality of independently temperature-controlled chambers, wherein, The incubator includes a refrigerant supply module and N chambers. The i-th chamber is equipped with a temperature sensor i, an evaporator i, a heating wire i, and an electronic expansion valve i. The refrigerant supply module is connected to the evaporator i via the electronic expansion valve i. The heating wire i provides heat energy to the chamber i, and the temperature sensor i is configured to acquire the sequential temperature within the chamber i. The artificial intelligence temperature control system includes N artificial intelligence subsystems that control the temperature of the N chambers respectively. The i-th artificial intelligence subsystem includes a reinforcement learning module i, a PID controller i, an electronic transfer switch i, and a subtractor i. The subtractor i is used to subtract the set temperature and the measured temperature within the chamber i to obtain an error signal. The feature is that the reinforcement learning module i, based on the error signal Generate the state of box i According to the status The control strategy of generating the PID controller i that controls the electronic expansion valve i or the heating wire i at time t. Specifically, it includes: S2-1: Calculate the state-action function at time t according to the following formula: ， where is the control parameter vector adopted by PID controller i at time t , the state is transitions to state with transition probability is the state value function at time t denotes the discount factor is the reward obtained by PID controller i adopting control parameter vector at time t S2-2: Improving according to the formula : ， In the formulae, is the time step; S2-3: Calculate the state value function at time t according to the following formula: S2-3: Calculate the state value function at time t according to the following formula: ； S2-4: judging whether or not it is less than or equal to a threshold value , if so, outputting a control parameter vector of the PID controller i at time ; if not, setting ; if not, setting , and then returning to step S2-1.

5. An artificial intelligence temperature control system, characterized by, This includes N artificial intelligence subsystems that control the temperature of N chambers respectively. The i-th artificial intelligence subsystem includes a reinforcement learning module i, a PID controller i, an electronic transfer switch i, and a subtractor i. The subtractor i is used to subtract the set temperature inside chamber i from the measured temperature to obtain an error signal. The reinforcement learning module i is based on the error signal Generate the state of box i According to the status The control strategy of generating the PID controller i that controls the electronic expansion valve i or the heating wire i at time t. ,in, In the formula, The control parameters of the PID controller i are the same at time t. These represent the proportional, integral, and derivative coefficients of the PID controller i at time t; the reinforcement learning module i also considers the error signal... The control strategy for generating electronic switching switch i at time t , i=1,…,N, This represents the control signal of electronic switch i at time t. wherein , wherein , and are the center and width of the jth Gaussian function of the neural network of reinforcement learning module i at time t; is the weight between the jth neuron of the hidden layer and the nth neuron of the output layer of the neural network of reinforcement learning module i at time t; j = 1, 2, 3; n = 1, 2, 3, denotes the two-norm.

6. The artificial intelligence temperature control system of claim 5, wherein, The reinforcement learning module i computes the state value function at time t according to : ， wherein is the weight between the jth neuron of the hidden layer of the neural network of the reinforcement learning module i and the Nth neuron of the output layer at time t, N = 4.

7. The artificial intelligence temperature control system of claim 6, wherein, The reinforcement learning module i builds a loss function according to the following formula : ， wherein Tset is the set temperature within the cabinet i, Tmeas is the measured temperature within the cabinet i; Update the parameters according to the following formula: ， ， ， In the formula, A control parameter vector was used for the PID controller i. And the rewards received; , , , The learning coefficient; To reinforce the learning module i, the time interval between the j-th neuron in the hidden layer of the neural network and the n-th neuron in the output layer is... The weights for each time period, n=1,2,3; To reinforce the learning module i, the time interval between the j-th neuron in the hidden layer of the neural network and the N-th neuron in the output layer is... The weights for each time period are N=4; To reinforce the learning module i, the time of the j-th Gaussian function in the neural network The center of time; To reinforce the learning module i, the time of the j-th Gaussian function in the neural network bandwidth at that time; Indicates to gradient, Indicates to The gradient; Expressed on The gradient; Indicates to The gradient; This is the discount factor; For time step.

Images