System and method for data-driven control of an air conditioning system

Abstract

A system for controlling the operation of an air conditioning system including a heat exchanger is provided. The system includes a processor executing a neural network trained to simulate operation of the heat exchanger in response to test control inputs and generate outputs of the simulation based on historical data defining the state of the heat exchanger. The historical data includes a series of historical control inputs provided to the heat exchanger and a series of historical outputs of operation of the heat exchanger corresponding to the series of historical control inputs. The processor determines control commands to the air conditioning system based on predicted test outputs of the simulation of operation of the heat exchanger in response to the test control inputs and transmits the determined control commands to actuators of the air conditioning system.

Description

[Technical Field]

[0001] The present disclosure relates generally to control systems, and more particularly to a control system and method for data-driven control of an air conditioner including a heat exchanger. [Background technology]

[0002] Control theory in control systems engineering is a subfield of mathematics that deals with the control of continuously operating dynamic systems in engineering processes and machines. Control systems perform control actions on dynamic systems with the aim of ensuring control stability.

[0003] For example, optimization-based control and estimation techniques such as model predictive control (MPC) enable a model-based design framework that can directly consider system dynamics and constraints. MPC is used in many applications to control dynamic systems of varying complexity. Examples of such systems include production lines, automobile engines, robots, numerically controlled machining, satellites, and power generators. However, in certain situations, analytical models of the controlled system may be unavailable, difficult to update in real time, or inaccurate. Examples of such cases are prevalent in the fields of robotics, heating, ventilation, and air conditioning (HVAC) systems, vehicles, smart grids, factory automation, transportation, self-tuning machines, and transportation networks.

[0004] For example, heat exchangers are widely used to transfer heat between air and working fluids in refrigeration, comfort cooling, and heating applications, and play a dominant role in the performance of heating, ventilation, air conditioning, and refrigeration (HVAC&R) equipment. However, the design and optimization of heat exchangers can be challenging due to numerous design variables, complex manufacturing constraints, and multiple conflicting objectives.

[0005] Generally, the dynamics of a heat exchanger can be described by conservation laws, i.e., by the balance of mass, energy, and momentum. Physics-based modeling approaches may require solving the resulting differential algebraic equations (DAEs) to predict the heat transfer and fluid flow dynamics occurring within the heat exchanger. However, DAE systems can be very difficult to solve, and challenges arise from index reduction, consistent initialization, and inherent rigidity. Furthermore, for large heat exchangers with complex circuits, the model complexity (number of equations) increases significantly. Because the temporal dynamics of HVAC systems can often be dominated by the temporal dynamics of the heat exchanger, simulation-based equipment design processes and other applications of predictive models are limited by the computational complexity of heat exchanger models. Summary of the Invention [Problem to be solved by the invention]

[0006] In the absence of accurate models of dynamic systems, some control methods utilize operational data generated by these systems to construct feedback control policies that stabilize system dynamics or embed quantifiable control-related performance. Using operational data for control is called data-driven control. Typically, there are two types of data-driven control methods, such as indirect and direct methods. In indirect methods, a model of the system is first constructed, and then a controller is designed utilizing this model. However, in direct methods, a control policy from operational data is constructed directly without an intermediate model-building step. A disadvantage of indirect methods is that the model-building stage can require a large amount of data. Conversely, direct methods require less data. However, even state-of-the-art direct control methods have difficulty addressing the state and input constraints essential for maintaining safe operation in closed-loop control systems.

[0007] Therefore, there is a need for an indirect, data-driven method for controlling one or more heat exchangers that reduces the amount of data for model building. [Means for solving the problem]

[0008] The present disclosure provides a control system and method for controlling the operation of an air conditioning system including a heat exchanger. The proposed control system and method can be utilized to generate a control policy for controlling the dynamic behavior of an air conditioning system including at least one heat exchanger using control actions in an optimal manner without delay or overshoot, and ensuring control stability of the air conditioning system.

[0009] It is an object of some embodiments to provide a control system and method for data-driven control of an air conditioning system including a heat exchanger. Additionally or alternatively, it is an object of some embodiments to provide a control system and method for indirect data-driven control using a neural network model derived from a realistic, predetermined window of historical data of the heat exchanger's operation.

[0010] Some embodiments are based on the recognition that the mechanics of a heat exchanger can be replaced by a digital twin that simulates the operation of the heat exchanger, outputting variables that a model of the heat exchanger (such as a neural network model) needs to estimate. The simulation of the operation of the heat exchanger by a digital twin model allows for a reduction in historical data compared to traditional indirect data-driven models of the heat exchanger.

[0011] Some embodiments are based on the recognition that conventional analytical models for air conditioning systems that include heat exchangers are based on physical conservation laws and require complex differential algebraic equations (DAEs) that are often difficult to solve. Furthermore, an increase in the number and size of heat exchangers in an air conditioning system leads to an increase in the complexity of the DAEs.

[0012] Some embodiments recognize that neural networks, such as convolutional recurrent networks, can accelerate the simulation of nonlinear dynamics in heat exchangers. Therefore, a deep state-space modeling framework is developed that combines the feature extraction capabilities of convolutional neural networks (CNNs) with the sequence prediction properties of gated recurrent units (GRUs). This modeling framework describes the fluid flow and heat transfer dynamics of heat exchangers in vapor compression cycles used in building energy systems.

[0013] Unlike modeling based on physical conservation laws, the proposed modeling approach develops relationships between input and output data without explicit knowledge of the underlying behavior and does not involve any DAEs. Furthermore, the number of model equations generated by the proposed modeling approach is independent of the heat exchanger's geometry, size, circuit pattern, and operating conditions. Therefore, the resulting model can be efficiently solved in real time, making it effective for online control of any air conditioning system that includes a heat exchanger.

[0014] To this end, some embodiments disclose a neural network comprising a combination of a convolutional network and a recurrent network. The neural network is trained to simulate the operation of a heat exchanger in response to test control inputs and to generate simulated outputs based on historical data defining the state of the heat exchanger. The historical data includes a series of historical control inputs to the heat exchanger and a series of historical outputs of the operation of the heat exchanger corresponding to the series of historical control inputs.

[0015] The neural network includes a first arm that processes boundary inputs appended to a series of historical boundary inputs using a first combination of a convolutional network and a recurrent network trained to extract features indicative of changes in the boundary inputs from the historical boundary inputs. The neural network further includes a second arm that processes a series of historical boundary inputs paired with a series of historical outputs using a second combination of a convolutional network and a recurrent network trained to extract output features indicative of dynamic coupling between the inputs and outputs of the heat exchanger's operation. The neural network further includes a third arm that processes the control features and output features to predict test outputs of the heat exchanger's operation corresponding to the boundary inputs.

[0016] Some embodiments disclose that the first and second arms of a neural network each include a bank of one-dimensional convolutional layers (e.g., CNNs) followed by a gated recurrent unit (GRU). In this way, the CNN-GRU leverages the concept of one-dimensional (1-D) convolutional networks for the first stage of analysis of time series data, which allows for better extraction of temporal features from inputs / outputs. The GRU is an efficient recurrent neural cell that analyzes the extracted temporal features to learn system dynamics. Furthermore, by splitting the data into an input time series (analyzed by the first arm) and its corresponding output (analyzed by the second arm), the neural network can independently learn the impact of input changes and the predicted output from the combination of the inputs and outputs in a dynamic system.

[0017] Some embodiments are based on the recognition that the third arm can be implemented as a fully connected deep neural network that accepts tensors containing control features and output features. The purpose of the fully connected deep neural network layer is to project the augmented GRU state into the output space.

[0018] Thus, the first arm analyzes the input time series, and the second arm analyzes the historical system output. In this way, the neural network understands the current state of the heat exchanger and / or air conditioning system because it knows the previous input and its corresponding output. With knowledge of the system dynamics, the current system state, and the next input, the neural network can accurately predict the output for the next time step.

[0019] Some embodiments disclose a control system having a controller that determines control commands to actuators of an air conditioning system based on the output of a predicted system model. Control commands can be sent to the air conditioning system to change compressor speed, open or close valves, or change the speed of a machine fan, etc.

[0020] Some embodiments are based on the recognition that different neural weights are assigned to the first arm path and the second arm path to increase the accuracy of the neural network.

[0021] Some embodiments are based on the recognition that a neural network can be trained with data trajectories generated by running a physics-based model of a heat exchanger having one or more boundary conditions associated with the heat exchanger. During training of the neural network, the data trajectories are divided into multiple batches. Each batch includes at least one of a set of historical control inputs, a set of historical outputs, an initial input to the machine, and a true output of the machine. Processing a predetermined window of data in a batch rather than the entire data trajectory reduces backpropagation time in training the GRU. Furthermore, avoiding the need to learn from the entire time series at once and instead using batches allows for scalability, parallelizability, and storage efficiency in the face of large datasets.

[0022] Some embodiments disclose that a mean squared error may be calculated for each batch of the multiple batches based on the predicted output of the heat exchanger neural network model and the true output of the air conditioning system, and a training loss associated with the calculated mean squared error for each batch is determined and optimized.

[0023] Some embodiments disclose that at each time step of a simulation, an air conditioning system model is initialized, including a room model, a compressor model, an expansion model, and a neural network model of a heat exchanger. Furthermore, boundary conditions on the air side of the air conditioning system, as well as compressor and expansion device settings, can be determined. Then, based on these conditions, the air conditioning system model is executed together with the room model. One or more states, including refrigerant states or air states associated with the room model and multiple component models, are generated. Processing each component model rather than a single model for the entire machine allows for a reduction in the data required to predict output data. Furthermore, processing a data-driven model with a similar neural network architecture for each heat exchanger is advantageous for controlling the dynamic behavior of an air conditioning system with multiple heat exchangers.

[0024] Some embodiments are based on the recognition that a plurality of test control inputs may be generated based on a Gaussian process. The plurality of test control inputs are then provided as inputs to a model of an air conditioning system to generate a plurality of predicted test outputs corresponding to the plurality of test control inputs. Control commands are determined based on the predicted test outputs of a simulation of the air conditioning system for the test control inputs selected based on optimizing a cost function associated with the plurality of test control inputs and the plurality of test outputs.

[0025] Some embodiments are based on the recognition that one or more test control inputs of a previous time point are utilized to generate multiple test control inputs based on a Gaussian process for a next time point following the previous time point.

[0026] Some embodiments are based on the recognition that a series of test control inputs corresponding to a finite number of time points can be generated. The series of test control inputs are provided as inputs to a model of the system to generate a series of predicted test outputs corresponding to the series of test control inputs. Next, a gradient of a cost function associated with the series of test control inputs and the series of test outputs is calculated. Furthermore, an optimal test control input is selected from the generated series of test control inputs based on the calculated gradient of the cost function, and control commands are determined based on predicted test outputs of a simulation of the operation of the air conditioning system for the selected test control input. Such a process allows for the selection of an optimal test control input for the model of the air conditioning system.

[0027] The present disclosure is further described in the following detailed description with reference to several drawings, in which like reference numerals represent like parts throughout the several views of the drawings, shown as non-limiting examples of exemplary embodiments of the present disclosure. The drawings shown are not necessarily to scale, emphasis instead generally being placed upon illustrating the principles of the presently disclosed embodiments. [Brief explanation of the drawings]

[0028] [Figure 1] FIG. 1 is a schematic diagram of an air conditioning system according to some embodiments of the present disclosure. [Figure 2A] FIG. 1 is a block diagram of a control system that utilizes a model to control an air conditioning system, according to some embodiments of the present disclosure. [Figure 2B] FIG. 1 is a block diagram of a neural network for predicting test outputs of a heat exchanger's operation corresponding to test control inputs in a single time step prediction, according to some embodiments of the present disclosure. [Figure 3A] FIG. 1 illustrates a detailed architecture of a neural network trained to simulate the operation of a heat exchanger included in an air conditioning system at a time step, according to some embodiments of the present disclosure. [Figure 3B]FIG. 1 illustrates a flowchart of an exemplary process for training a neural network, according to some embodiments of the present disclosure. [Figure 3C] FIG. 1 illustrates a flowchart for optimizing training loss for training neural networks, according to some embodiments of the present disclosure. [Figure 4] FIG. 1 is an exemplary diagram of a model of an air conditioning system, according to some embodiments of the present disclosure. [Figure 5] FIG. 1 illustrates a flowchart for generating a room model and one or more states associated with multiple component models of an air conditioning system according to some embodiments of the present disclosure. [Figure 6] FIG. 1 is a block diagram illustrating various components of a control system for controlling an air conditioning system, according to some embodiments of the present disclosure. [Figure 7A] FIG. 1 illustrates a flowchart of an exemplary method for selecting an optimal test control input from multiple test control inputs based on a Gaussian process, according to some embodiments of the present disclosure. [Figure 7B] FIG. 10 shows a flowchart illustrating the generation of multiple test control inputs according to some embodiments of the present disclosure. [Figure 8] FIG. 1 illustrates a flowchart of an exemplary method for selecting optimal test control inputs from a set of test control inputs based on a model predictive control process, according to some embodiments of the present disclosure. [Figure 9] FIG. 1 shows a flowchart illustrating a method for determining control commands to an air conditioning system according to some embodiments of the present disclosure. [Figure 10] FIG. 1 is a block diagram of a use case for controlling an air conditioning system using a control system, according to some embodiments of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0029] In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. However, it will be apparent to those skilled in the art that the present disclosure may be practiced without these specific details. In other instances, devices and methods are shown in block diagram form solely to avoid obscuring the disclosure. It is contemplated that various changes may be made to the function and arrangement of elements without departing from the spirit and scope of the disclosed subject matter as set forth in the appended claims.

[0030] As used in this specification and claims, the terms "for example," "for example," and "such as," as well as the verbs "comprise," "have," and "include," and each of their other verb forms, when used in conjunction with a list of one or more components or other items, should be construed as open-ended, meaning that the list should not be considered as excluding other additional components or items. The term "based on" means based at least in part on. Furthermore, it should be understood that the style and terminology used herein are for purposes of description and should not be regarded as limiting. Any headings used herein are for convenience only and have no legal or limiting effect.

[0031] Specific details are provided in the following description to provide a thorough understanding of the embodiments. However, those skilled in the art will understand that the embodiments may be practiced without these specific details. For example, systems, processes, and other elements of the disclosed subject matter may be shown as components in block diagram form in order to avoid obscuring the embodiments in unnecessary detail. In other instances, well-known processes, structures, and techniques may be shown without unnecessary detail in order to avoid obscuring the embodiments. Furthermore, like reference numbers and names in the various drawings indicate like elements.

[0032] In describing the disclosed embodiments, the following definitions apply throughout this disclosure.

[0033] A "control system" or "controller" may refer to a device or set of devices for managing, directing, supervising, or regulating the behavior of other devices or systems. A control system may be implemented in either software or hardware and may include one or more modules. Control systems that include feedback loops may be implemented using a microprocessor. A control system may also be an embedded system.

[0034] "Vapor compression system" refers to a system that uses a vapor compression cycle to pass a refrigerant through system components based on principles of thermodynamics, fluid mechanics, and / or heat transfer. Vapor compression systems can be, but are not limited to, heat pumps, refrigeration, and air conditioning systems. Vapor compression systems are used in applications beyond conditioning residential or commercial spaces. For example, vapor compression cycles may be used to cool computer chips in high-performance computing applications.

[0035] "Vapor compression system component" refers to any component of a vapor compression system whose operation is controllable by the control system, including, but not limited to, a variable speed compressor for compressing and delivering refrigerant to the system, an expansion valve for providing an adjustable pressure drop between high and low pressure portions of the system, and evaporative and condensing heat exchangers, each containing a variable speed fan for regulating the airflow through the heat exchanger.

[0036] "Evaporator" refers to a heat exchanger in a vapor compression system in which refrigerant passing through it evaporates as it passes through the heat exchanger, resulting in a higher specific enthalpy of the refrigerant at the outlet of the heat exchanger than at the inlet, and the refrigerant typically changing from a liquid to a gas. There may be one or more evaporators in a vapor compression system.

[0037] "Condenser" refers to a heat exchanger in a vapor compression system in which refrigerant passing through the heat exchanger condenses as it passes through the heat exchanger, resulting in a lower specific enthalpy of the refrigerant at the outlet than at the inlet, and the refrigerant typically changing from a gas to a liquid. One or more condensers may be in a vapor compression system.

[0038] A "refrigerant circuit" refers to the interconnection of refrigerant pipes and components into a closed configuration, so that refrigerant flows in a closed path between a series of components. Refrigerant circuits can be used to create closed thermodynamic cycles for efficiently transferring thermal energy from one location to another. For example, the refrigerant circuit of a vapor compression system includes a compressor, a condensing heat exchanger, an expansion valve, an evaporating heat exchanger, and pipes used to transport the refrigerant from each component to the next.

[0039] "Electrical circuitry" refers to interconnections of wires that transmit electrical signals between components such as processors, memories, or actuators.

[0040] A "set of control signals" refers to specific values ​​of inputs for controlling the operation of components of a vapor compression system. The set of control signals includes, but is not limited to, a compressor speed value, an expansion valve position value, an evaporator fan speed value, and a condenser fan speed value.

[0041] "Setpoint" refers to a target value that a system, such as a vapor compression system, aims to reach and maintain as a result of operation. The term setpoint applies to a particular set of control signals and any particular values ​​of thermodynamic and environmental parameters.

[0042] "Computer" refers to any device capable of accepting structured input, processing the structured input according to predetermined rules, and producing the results of the processing as output. Examples of computers include general-purpose computers, supercomputers, mainframes, superminicomputers, minicomputers, workstations, microcomputers, servers, interactive televisions, hybrid combinations of computers and interactive televisions, and special-purpose hardware for emulating computers and / or software. A computer may have a single processor or multiple processors that may or may not operate in parallel. A computer also refers to two or more computers connected to each other via a network to send or receive information from one another. An example of such a computer is a distributed computer system for processing information via computers linked by a network.

[0043] "Central Processing Unit (CPU)" or "Processor" refers to a computer or a component of a computer that reads and executes software instructions.

[0044] "Memory" or "computer-readable medium" refers to any storage device for storing data accessible by a computer. Examples include magnetic hard disks, floppy disks, optical disks such as CD-ROMs or DVDs, magnetic tape, memory chips, carrier waves used to carry computer-readable electronic data such as those used in sending and receiving email or accessing a network, and computer memory, e.g., random access memory (RAM).

[0045] "Software" refers to predetermined rules for operating a computer. Examples of software include software, code segments, instructions, computer programs, and programmed logic. Software in intelligent systems can be self-learning.

[0046] A "module" or "unit" refers to a basic component within a computer that performs a task or part of a task. This may be implemented in either software or hardware.

[0047] "Controller," "control system," and / or "regulator" refer to a device or set of devices for managing, directing, supervising, or regulating the behavior of other devices or systems. A controller may be implemented by hardware, a processor with software-configured operations, and combinations thereof. A controller may also be an embedded system.

[0048] FIG. 1 illustrates a schematic diagram 100 of an air conditioning system 102 according to an embodiment of the present disclosure. The schematic diagram 100 includes a space 104 and an air conditioning system 102 configured to provide heating or cooling to the space 104. The air conditioning system 102 may include various components, such as variable setting actuators for performing operations such as a refrigerant cycle. For example, the air conditioning system 102 may include a variable speed compression device 106, an outdoor heat exchanger 108, a variable speed outdoor fan 110, an expansion device 112, an indoor heat exchanger 114, and an air circulation blower 116. The various components may be connected in a closed-loop serial refrigerant flow arrangement. The air conditioning system 102 may function as a heating or cooling device and may provide thermal comfort to occupants by adjusting the required capacity to match the load in the space 104. Air is conditioned by the indoor heat exchanger 114 and delivered to the space 104 by the air circulation blower fan 116. Depending on the operating mode of the air conditioning system 102 , after absorbing or removing heat from the space 104 , the air may be circulated back to the indoor heat exchanger 114 .

[0049] In general, the temporal dynamics of the air conditioning system 102 depend heavily on the dynamics of the heat exchangers of the air conditioning system 102. The heat exchangers of the air conditioning system 102 may be the indoor heat exchanger 114 or the outdoor heat exchanger 108. The dynamics of the air conditioning system 102 may be described by conservation laws, i.e., by the balance of mass, energy, and momentum. Physics-based analytical models of heat exchangers require solving complex differential algebraic equations (DAEs) to predict the heat transfer and fluid flow dynamics occurring within the heat exchanger. Furthermore, for large heat exchangers with complex circuits, the complexity of the physics-based analytical models may increase significantly.

[0050] An objective of some embodiments is to disclose an apparatus for data-driven control of a system including a heat exchanger, such as the air conditioning system 102. Another objective of some embodiments is to disclose a method for data-driven control of a system including a heat exchanger. The proposed apparatus of the present disclosure may include a control system for data-driven control of the dynamic behavior of the air conditioning system 102 including a heat exchanger. The control system and method proposed in this disclosure eliminate the need to solve these complex DAEs. Typically, a single data-driven model of the entire air conditioning system 102 requires a large amount of data to predict outputs. To collect this large amount of data, a physics-based model of the air conditioning system 102 must be run with various inputs, which is tedious and can increase the likelihood of error. Furthermore, data-driven system models are not reusable even with slight changes to the system architecture unless additional inputs are included to accommodate those changes, which inevitably increases the dimensionality of the model. Instead, a module-based solution approach is more preferable. Specifically, at the component level, data-driven models are used to fully or partially replace physics-based component models. At the system level, this follows the first principle of satisfying the conservation of mass, energy, and momentum. Therefore, controlling data-driven models for the components of the air conditioning system 102 according to the present disclosure allows for a reduction in data requirements and eliminates the complexity of model training. If any component in the air conditioning system 102 is changed or added, the data-driven model for that specific component only needs to be regenerated, while other models remain unchanged. Furthermore, the number of model equations generated by the data-driven model is independent of the geometry, size, circuit pattern, and operating conditions of components such as heat exchangers. Therefore, data-driven models are advantageous for online control of any machine, including heat exchangers, because they can be solved efficiently in real time.

[0051] It will be understood that this disclosure considers the case of an air conditioning system for use with data-driven control. However, such data-driven control may equally be used in any machine that includes at least one heat exchanger. Examples of machines include, but are not limited to, vehicle systems, smart grid systems, factory automation systems, transportation systems, self-tuning machine systems, and traffic networks.

[0052] 2A shows a block diagram 200A of a control system 202 utilizing a model 204 to control an air conditioning system 102, according to an embodiment of the present disclosure. The block diagram 200A may further include a system 206. The system 206 includes the air conditioning system 102 and the space 104.

[0053] In one embodiment, control system 202, model 204, and system 206 communicate via a network (not shown in FIG. 2A ). The network may include a public network such as the Internet, a telephone network, a satellite network, and / or a local area network (LAN) such as Ethernet, a wide area network (WAN). In addition, examples of networks may include dedicated line networks such as Internet Protocol Virtual Private Networks (IP-VPN). In addition, the network may include wireless communication networks such as Wi-Fi, Bluetooth, etc.

[0054] The control system 202 is configured to control the functions of the system 206. The control system 202 includes at least one processor, such as processor 202A, configured to execute computer-executable instructions. The control system 202 further includes memory 202B that stores computer-executable instructions executable by the processor 202A. The processor 202A may be a single-core processor, such as a central processing unit (CPU), a multi-core processor, a computing cluster, or any number of other configurations. The memory 202B may include non-transitory memory, such as random access memory (RAM), read-only memory (ROM), flash memory, or any other suitable memory system. The processor 202A is configured to control the functions of all components of the control system 202. Furthermore, the control system 202 is configured to generate actuator commands to control the system 206. Additionally, the control system 202 controls the model 204 due to execution and training of the model 204 by the processor 202A.

[0055] The model 204 includes a simulation model of the air conditioning system 102 and the space 104. Test control inputs are sent to the model 204 by the control system 202. The model 204 includes a heat exchanger model having a neural network 208 configured to simulate the operation of the heat exchanger corresponding to the test control inputs at each time step of a multi-stage prediction, as described below with reference to FIG. 2B. The data-driven heat exchanger model allows the system model 204 of the air conditioning system 102 to be more computationally efficient by eliminating the need to solve complex DAEs. Details of the neural network heat exchanger model are further provided with reference to FIGS. 2B and 3A.

[0056] The control system 202 may receive a reference signal 210 (also denoted “r”) and measurement data 212 (also denoted “v”) from the system 206 and generate an actuator command output 214 (also denoted “u”) to control the operation of the system 206. The actuator command output 214 is optimized to minimize a cost function while satisfying the reference signal 210. In one embodiment, the cost function may be the power consumption of the system 206. In another embodiment, the cost function may be the tracking error between the reference signal 210 and the measurement data 212. In one embodiment, the reference signal 210 may be related to, for example, the air temperature of the space 104 and the inlet superheat of the air conditioning system 102. In another embodiment, the reference signal 210 may be related to the air temperature of the space 104 and the discharge temperature of a component of the air conditioning system 102. The actuator command output 214 may be related to at least one of a compressor speed, a valve opening, or a fan speed of the air conditioning system 102.

[0057] Processor 202A of control system 202 generates test control inputs 216 and sends them to model 204 of system 206. Unlike first principles models of vapor compression systems, which are built on conservation laws and are very computationally expensive due to the resulting DAEs, model 204 consists of a very fast data-driven component model, allowing for many simulations in a short amount of time. Processor 202A is further configured to execute model 204 based on historical operating data 218 from system 206 and test control inputs 216 from control system 202 to obtain corresponding test outputs 220. Control system 202 may evaluate test control inputs 216 based on various metrics and determine actuator command outputs 214(u) for system 206.

[0058] FIG. 2B shows a block diagram 200B of a neural network 208 for predicting test outputs 222 of the operation of a heat exchanger corresponding to test control inputs 224 in a single time step prediction.

[0059] The neural network 208 includes a first arm 226 having a first combination of convolutional networks and recurrent networks. The first combination of convolutional networks and recurrent networks includes a bank of one-dimensional (1-D) convolutional neural networks (CNNs) 228 and a recurrent neural network (RNN) 230. The neural network 208 further includes a second arm 232 having a second combination of convolutional networks and recurrent networks. The second combination of convolutional networks and recurrent networks includes a bank of 1-D CNNs 234 and an RNN 236. The bank of 1-D CNNs 228 of the first arm 226 is provided with a test control input 224 and a set of historical control inputs 238 to extract control features 240. The control features 240 indicate changes in the test control input 224 from the set of historical control inputs 238. The first arm 226 is configured to process the test control inputs 224 appended to a set of history control inputs 238 using a first combination of a bank of 1-D CNNs 228 and an RNN 230. The bank of 1-D CNNs 234 of the second arm 232 is provided with a set of history outputs 242 paired with the set of history inputs 238 to extract output features 244. The output features 244 indicate a dynamic coupling between the test control inputs 224 and the corresponding test outputs 222 of the heat exchanger operation. The second arm 232 is configured to process the set of history inputs 238 paired with the set of history outputs 242 using a second combination of the bank of 1-D CNNs 234 and an RNN 236. The neural network 208 further includes a third arm 246 configured to process the control features 240 and the output features 244 to predict the test outputs 222 of the heat exchanger operation corresponding to the test control inputs 224.

[0060] 3A illustrates an architecture 300A of a neural network 208 trained to simulate the operation of a heat exchanger included in an air conditioning system 102 at time step t, according to an embodiment of the present disclosure. For example, the neural network 208 may be trained to simulate the operation of a heat exchanger.

[0061]

number

[0062] The set of boundary conditions X consists of inlet conditions for each fluid stream, such as mass flow rate, pressure, and specific enthalpy or other thermodynamic variables. The initial state variables x0 represent the initial values ​​of the neural network 208 differential variables, such as the pressure and specific enthalpy or tube wall temperature of the fluid in each control volume. Finally, the set of output variables Y may consist of variables defined at each volume, such as the air temperature or moisture content leaving each volume, or may consist of aggregate variables for the entire heat exchanger, such as the total heat power transferred from one fluid to another.

[0063] A deep state space modeling framework that combines the feature extraction capabilities of CNNs with the sequence prediction properties of RNNs is incorporated into the neural network 208.

[0064] According to some embodiments, gated recurrent units (GRUs) may be used here as RNNs 230 and / or 236. GRUs are efficient recurrent neural cells configured to analyze extracted temporal features to learn system dynamics. According to some embodiments, bank of 1-D CNNs 228 and bank of 1-D CNNs 234 are used here for the first stage of analysis of time series data. Therefore, the architecture of neural network 208 may be referred to as a CNN-GRU deep state-space model (SSM). A CNN-GRU SSM is a neural network that generates a set of historical control inputs (X t-N:t-1) 238 and a series of historical control inputs (X t-N:t-1 )238 corresponding series of historical outputs (Y t-N:t-1 ) 242. t ) to find the test output (Y t ) 222. Further, referring to FIG. 2B, a prediction of the set of historical control inputs (X t-N:t-1 )238 and a series of history outputs (Y t-N:t-1 ) 242 constitutes part of the operational data 218, and the test control inputs (X t ) 224 constitutes part of the test control input 216 .

[0065] Additionally, processor 202A (shown in FIG. 2A) processes the windowed data sequence through two paths: X for the upper X path 302; t-N:t and for the lower (X;Y) path 304, t-N:t-1 and Y t-N:t-1 In the upper X path 302 and the lower (X;Y) path 304, the windowed data sequence passes through the CNN-GRU SSMs of the first arm 226 and the second arm 232, respectively. The windowed data then flows through the bank of 1-D CNNs 228 and the bank of 1-D CNNs 234, respectively, for feature extraction. The upper X path 302 is configured to generate a combination of the test control input (X t ) 224 and seeks to learn the impact of previously observed changes in the inputs on the predicted output. The lower (X;Y) path 304 extracts features relevant to learning how the inputs and outputs are connected in a dynamical system. Furthermore, both the upper X path 302 and the lower (X;Y) path 304 continue to their respective RNNs 230 and 236, which model how the internal hidden states perform updates based on the features extracted from the bank of 1-D CNNs 228 and bank of 1-D CNNs 234, respectively.

[0066]

number

[0067] At time t, the internal state of the upper X path 302, which is the extracted control feature 240, is denoted by "h", while the internal state of the lower (X;Y) path 304, which is the extracted output feature 244, is denoted by "d". Thus, the internal state of the RNN 230 is "h" 240, and the internal state of the RNN 236 is "d" 244. The internal states h 240 and d 244 are concatenated into a single augmented state vector and passed to a final set of third arms 246. The third arms 246 include a fully connected layer where every input neuron is connected to every output neuron. The purpose of the fully connected layer is to project the augmented RNN state into the output space. The augmented state vector may be a tensor that is accepted by the third arms 246. The output of the third arms 246 is the predicted test output Y t The number is 222.

[0068] Next, the predicted output Y t 222 is passed to the next time step 306 and used to generate (X;Y) data 308 for the next time step.

[0069] The architecture of the proposed neural network 208 is designed to promote several specific beneficial properties. For example, adding a bank of 1-D CNN layers facilitates temporal feature extraction and preprocessing to aid the GRU cells.

[0070]

number

[0071]

number

[0072] Additionally, at block 314, training of the neural network 208 is completed and the refrigerant condition and air condition are received as outputs.

[0073] 3C shows a flowchart 300C for optimizing the training loss for training the neural network 208, according to an embodiment of the present disclosure. Flowchart 300C specifically shows an expansion of step 312 of FIG. 3B.

[0074]

number

[0075] Considering a window of length N rather than the entire dataset ultimately reduces the training time of the GRU cell, as the unwrapping of the temporal backpropagation required to train the GRU is not prohibitively expensive. Aside from time efficiency, avoiding the need to learn from the entire time series at once and instead using batching enables scalability, parallelizability, and storage efficiency in the face of large datasets. Additionally, by forcing the algorithm to treat each windowed time series segment independently from other windows (including consecutive windows), the proposed deep SSM learns one-step updates of dynamics that are independent of the starting time, with the time dependence embedded in the GRU cell's internal state vector. This style of modeling is often advantageous over autoregressive approaches that use neural SSM predictions to forecast the future, since recursive estimation can often lead to error accumulation and degradation of model performance.

[0076]

number

[0077]

number

[0078]

number

[0079] FIG. 4 illustrates an example diagram 400 of a model 204 according to an embodiment of the present disclosure. The model 204 may include a room model 402A of a room 402 and multiple models corresponding to components of the air conditioning system 102. Examples of components of the air conditioning system 102 may include a compressor 404, a condenser 406, an expansion device 408, and an evaporator 410. Examples of multiple component models may include a performance map-based compressor model 404A of the compressor 404, a neural network 208 of the condenser 406, a performance map-based expansion device model 408A of the expansion device 408, and a neural network 208 of the evaporator 410. Furthermore, the room model 402A may be a lumped parameter room model. In a lumped parameter model, spatial variations in parameters can be ignored, and the system can be described by adjustable parameters. Furthermore, the room 402 includes a space 104 to which the air conditioning system 102 is configured to provide heating or cooling.

[0080] Performance map-based models map the performance of a component based on experimental data corresponding to the component. One example of a performance map-based compressor model is the ARI 10-coefficient compressor model, which calculates the compressor mass flow rate and compressor power output based on saturated suction and discharge temperatures. Other types of performance map-based compressor models determine the mass flow rate and power consumption based on suction and discharge pressures. In performance map-based expansion valve models, these models typically calculate the mass flow rate through the valve based on the pressure difference across the valve, where the flow coefficient is a function of the valve opening.

[0081]

number

[0082] All component models implement the same standard interface so that model 204 can handle any system configuration and employ a component-based solution scheme. Thus, the outlet boundary conditions of an upstream model are the inlet boundary conditions of a downstream model. Specifically, compressor model 404A is an upstream model of condenser model 208. Condenser model 208 is an upstream model of expansion device model 408A. Expansion device model 408A is an upstream model of evaporator model 208. Evaporator model 208 is an upstream model of compressor model 404A. Similarly, condenser model 208 is a downstream model of compressor model 404A. Expansion device model 408A is a downstream model of condenser model 208. Evaporator model 208 is a downstream model of expansion device model 408A. Compressor model 404A is a downstream model of evaporator model 208.

[0083]

number

[0084] 5 illustrates a flowchart 500 for generating one or more states associated with a room model 402A and multiple component models of an air conditioning system 102, according to an embodiment of the present disclosure. In block 502, the processor 202A is configured to initialize a model 204 of the system 206 and determine one or more conditions associated with the air side and the refrigerant side of a heat exchanger. Parameters of the air side conditions of the heat exchanger are air mass flow rate, air temperature, and relative humidity. Initial conditions of the refrigerant side of the heat exchanger are a refrigerant pressure distribution and a refrigerant enthalpy distribution. Based on the initial conditions and boundary conditions, the model 204 of the system 206 is fully defined.

[0085] At block 504, a set of historical control inputs and a set of historical outputs of the operation of the heat exchanger are provided as inputs to the neural network 208 of the heat exchanger model.

[0086] Additionally, in block 506, the processor 202A executes all component models, such as the performance map based compressor model 404A, the neural network 208 of the condenser 406, the performance map based expansion device model 408A, and the neural network 208 of the evaporator 410, along with the room model 402A. Because the refrigerant flow conditions at interfaces 412, 414, 416, and 418 of the system model 204 and the air flow conditions at interface 420 are iterated variables, these models are mathematically closed form and can be executed independently.

[0087] In block 508, processor 202A is configured to collect the outputs obtained in block 506 from each component model of air conditioning system 102. System solver 422 then formulates residual equations at each interface, such as interface 412, interface 414, interface 416, and interface 418, to ensure conservation of mass, energy, and momentum.

[0088] At block 510, the processor 202A is configured to execute the system solver 422, which determines whether convergence criteria based on the conservation of mass, energy, and momentum are met at each interface.

[0089] At block 512, the processor 202A generates one or more states associated with each of the room model and the plurality of component models. The one or more states include at least one of a refrigerant state or an air state associated with the air conditioning system 102. Furthermore, the one or more states associated with each model are obtained based on the convergence criterion determination at block 510 and are used as initial conditions for the calculation of the next time step.

[0090] At block 514, the process of setting the inputs to the neural network 208 is repeated at each time step until the entire heat exchanger simulation is completed.

[0091] 6 shows a block diagram 600 illustrating various components of a control system 202 for controlling an air conditioning system 102, according to an embodiment of the present disclosure. The components of the system 206 include the compressor 404, a condenser fan 602 for the condenser 406, an expansion valve 604, and an evaporator fan 606 for the evaporator 410. An electrical circuit 608 includes a condenser fan controller 610 for controlling one or more parameters of the condenser fan 604, a compressor controller 612 for controlling one or more parameters of the compressor 404, an expansion valve controller 614 for controlling one or more parameters of the expansion valve 604, and an evaporator fan controller 616 for controlling one or more parameters of the evaporator fan 606. For example, controllers such as condenser fan controller 610, compressor controller 612, expansion valve controller 614, and evaporator fan controller 616 may control, but are not limited to, the speed of the condenser fan 602 and the evaporator fan 606, the threshold, ratio, knee, attack time, release time, or makeup gain of the compressor 404, the valve body, diaphragm, pin or needle, spring, sensing valve, and capillary line of the expansion valve 604. Furthermore, the controllers of the electrical circuitry 608 may execute control processes to control the components of the system 206 based on the actuator command outputs 214 received from the control system 202. In addition, the components of the system 206 generate corresponding outputs in response to the control processes executed by the electrical circuitry 608 based on the actuator command outputs 214. Each component of the air conditioning system 102 of the system 206 is equipped with one or more sensors, collectively referred to herein as sensors 618. The sensor 618 records the measurement data 212 of the air conditioning system 102 and sends it back to the control system 202. The control system 202 controls the simulation of the model 204 to reduce the error between the reference signal 210 and the measurement data 212.

[0092] FIG. 7A illustrates a flowchart 700A of an exemplary method for selecting an optimal test control input from test control inputs 216 based on a Gaussian process, according to an embodiment of the present disclosure. In this embodiment, at the current time, the control system 202 generates an optimal control input for the next time step. To that end, in block 702, the processor 202A is configured to generate the test control input 216 based on a Gaussian process. In block 704, the test control input 216 is sent to the model 204. In block 706, the processor 202A receives a plurality of test outputs 220 predicted by the model 204 corresponding to the test control input 216. To that end, a simulation of the model 204 is performed for each of the test control inputs 216 to predict the corresponding test output 220 for the next time step. The processor 202A calculates a cost function for each of the test control inputs 216 with respect to the corresponding test output 220 and the reference signal 210. Examples of cost functions are described above with reference to FIG. 2B.

[0093] At block 708, the processor 202A is further configured to select an optimal test control input from the test control inputs 216 based on optimizing a cost function associated with the test control inputs 216 and the test outputs 220. The selected optimal test control input thus optimizes the cost function and is selected as the actuator command output 214 for the next time step. The processor 202A further determines a control command based on the selected optimal test control input. The control command may be sent to the air conditioning system 102 to change the speed of the compressor 404, open or close the expansion valve 604, change the speed of the condenser fan 602, and / or change the speed of the evaporator fan 606, etc.

[0094] 7B shows a flowchart 700B illustrating the generation of test control inputs 216, according to an embodiment of the present disclosure. The test control inputs 216 are generated based on one-step optimization-based control.

[0095] Initially, the control system 202 may consider a uniform distribution of test control inputs because no information about test control inputs from previous time steps is available. These uniform test control inputs are sent to the system model 204. The processor 202A is configured to run a simulation to evaluate each of these uniform test control inputs. The control input that minimizes the cost function is selected as the actuator command 214 for the first time step. The processor 202A further utilizes one or more test control inputs from previous time points to generate multiple test control inputs 216 based on a Gaussian process for the next time point. To this end, in block 710, the control system 202 collects test control inputs from previous time steps. In block 712, the processor 202A is configured to generate the test control inputs 216 for the current time step based on the test control inputs from the previous time step. The processor 202A may generate the test control inputs 216 based on the Gaussian process of FIG. 7A . To this end, the mean and variance of the Gaussian process are calculated. For example, the mean of the generated test control input 216 should be the control input at the previous time step, while the variance can be found as follows:

[0096] Given all test control inputs at the previous time step, the corresponding predicted outputs are available because these predicted outputs are based on the execution of the model 204 at the previous time step.

[0097] Calculate the mean and variance of all predicted outputs at the previous time step, discard outputs that are outside two standard deviations from the mean, and keep the remaining outputs.

[0098] The variance σ of the test control input 216 corresponding to the predicted output retained in step (2) t-1 Calculate.

[0099] The variance of the Gaussian distribution is σ t-1is used to generate the test control input 216 at the current time step.

[0100] 8 illustrates a flowchart 800 of an exemplary method for selecting an optimal test control input from a set of test control inputs based on a model predictive control process, according to another embodiment of the present disclosure. In this embodiment, the test control input 216 may be generated based on nonlinear model predictive control. Unlike one-step optimization-based control, gradient-based control evaluates a cost function over a finite time horizon rather than at a single point in time.

[0101] In block 802, the processor 202A generates an initial guess for a set of test control inputs 216 corresponding to a finite number of time points. Next, in block 804, the processor 202A calculates the gradient of the cost function with respect to the set of test control inputs 216 based on the predicted outputs 220 from the model 204. To that end, the set of test control inputs 216 are provided as inputs to the model 204, and a set of test outputs 220 corresponding to the set of test control inputs 216 are predicted by the model 204. Furthermore, the processor 202A uses numerical differentiation to calculate the gradient of the cost function with respect to the set of test control inputs 216. Because the model 204 is very computationally efficient, numerical differentiation does not impose a significant overhead during this process. In block 806, the set of test control inputs 216 is adjusted along the gradient of the cost function.

[0102] Further, the processor 202A selects an optimal test control input from the generated set of test control inputs 216 based on the gradient of the calculated cost function. To that end, in block 808, the processor 202A compares the gradient of the cost function for the set of test control inputs 216 with a predetermined tolerance. Then, in block 810, if the gradient of the cost function associated with the set of test control inputs 216 is less than the predetermined tolerance, the first step in the set of test control inputs 216 is selected as the optimal control input. If the gradient of the cost function is greater than the predetermined tolerance, the process in block 806 is repeated until the entire process converges. The processor 202A executes the model 204 to predict test outputs based on the test control inputs and determines control commands for the air conditioning system 102 based on the predicted test outputs of the model 204.

[0103] FIG. 9 shows a flow chart 900 illustrating a method for determining control commands to the air conditioning system 102 .

[0104] In block 902, the processor 202A determines control commands for the air conditioning system 102 based on predicted test outputs 222 of a simulation of the heat exchanger's operation in response to test control inputs 224. To do so, the model 204 predicts test outputs 220 based on the predicted test outputs 222 of the neural network 208. Using the predicted test outputs 220 of the model 204, the reference signals 210, and the corresponding measurement data 212, the processor 202A of the control system 202 generates actuator command outputs 214. The processor 202A determines control commands for the air conditioning system 102 based on the generated actuator command outputs 214. Examples of control commands are described with respect to FIG. 7A.

[0105] In block 904, the processor 202A sends control commands to actuators such as the condenser fan controller 610, the compressor controller 612, the expansion valve controller 614, and the evaporator fan controller 616 of the air conditioning system 102.

[0106] FIG. 10 shows a block diagram 1000 for controlling an air conditioning system 102 using a control system 202 according to an embodiment of the present disclosure.

[0107] The air conditioning system 102 is disposed in a space 104. The space 104 is occupied by occupants 1002, 1004, 1006, and 1008. Arrow 1010 represents air supplied to the space 104 by the air conditioning system 102. The control system 202 executes the model 204 and receives the test output 220.

[0108] Furthermore, using the test output 220, the reference signal 210, and the measurement data 212 of the system 206, the control system 202 generates control commands 214 for the system 206. The control commands further control the components of the air conditioning system 102 in an optimal manner.

[0109] The above description provides exemplary embodiments only and is not intended to limit the scope, applicability, or configuration of the disclosure. Rather, the following description of exemplary embodiments will provide those skilled in the art with an enabling description for implementing one or more exemplary embodiments. Various changes are contemplated that may be made in the function and arrangement of elements without departing from the spirit and scope of the disclosed subject matter as set forth in the appended claims.

[0110] Specific details are provided in the following description to provide a thorough understanding of the embodiments. However, it will be understood by those skilled in the art that the embodiments may be practiced without these specific details. For example, systems, processes, and other elements of the disclosed subject matter may be shown as components in block diagram form in order to avoid obscuring the embodiments in unnecessary detail. In other instances, well-known processes, structures, and techniques may be shown without unnecessary detail in order to avoid obscuring the embodiments. Furthermore, like reference numbers and names in the various drawings indicate like elements.

[0111] Also, particular embodiments may be described as a process that is depicted as a flowchart, a flow diagram, a data flow diagram, a structure diagram, or a block diagram. While a flowchart may describe operations as a sequential process, many of the operations may be performed in parallel or simultaneously. Additionally, the order of operations may be rearranged. A process may be terminated when its operations are completed, but may have additional steps not discussed or included in the diagram. Moreover, not all operations in any specifically described process may occur in all embodiments. A process may correspond to a method, a function, a procedure, a subroutine, a subprogram, etc. When a process corresponds to a function, the termination of the function may correspond to a return of the function to the calling function or the main function.

[0112] Furthermore, embodiments of the disclosed subject matter may be implemented, at least in part, either manually or automatically. The manual or automatic implementation may be performed or at least assisted by machine, hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When implemented in software, firmware, middleware, or microcode, the program code or code segments to perform the necessary tasks may be stored on a machine-readable medium. A processor(s) may perform the necessary tasks.

[0113] The various methods or processes outlined herein may be coded as software executable on one or more processors using any one of a variety of operating systems or platforms. Additionally, such software may be written using any of a number of suitable programming languages ​​and / or programming or scripting tools, and may be compiled as executable machine language code or intermediate code that runs on a framework or virtual machine. Typically, the functionality of the program modules may be combined or distributed as desired in various embodiments.

[0114] Each embodiment is described as a process, which is depicted as a flowchart, flow diagram, data flow diagram, structure diagram, or block diagram. While a flowchart depicts operations as a sequential process, many of the operations can be performed in parallel or simultaneously. Additionally, the order of operations may be rearranged. A process may be terminated when its operations are completed, but may have additional steps not discussed or included in the diagram. Moreover, not all operations in any process specifically described may occur in all embodiments. A process may correspond to a method, a function, a procedure, a subroutine, a subprogram, etc. When a process corresponds to a function, the termination of the function may correspond to a return of the function to the calling function or the main function.

[0115] Furthermore, embodiments of the disclosed subject matter may be implemented, at least in part, either manually or automatically. The manual or automatic implementation may be performed or at least assisted by machine, hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof. When implemented in software, firmware, middleware, or microcode, the program code or code segments to perform the necessary tasks may be stored on a machine-readable medium. A processor(s) may perform the necessary tasks.

[0116] Numerous modifications and other embodiments of the disclosures described herein will come to mind to one skilled in the art to which the disclosures described herein pertain having the benefit of the teachings presented in the above description and the associated drawings. It should be understood that the disclosure is not limited to the particular embodiments disclosed, and that modifications and other embodiments are intended to be included within the scope of the appended claims. Moreover, while the above description and the associated drawings describe exemplary embodiments in the context of certain illustrative combinations of elements and / or functions, it should be recognized that alternative embodiments may provide different combinations of elements and / or functions without departing from the scope of the appended claims. In this regard, combinations of elements and / or functions other than those expressly described above are also contemplated, for example, as may be set forth in some of the appended claims. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.

Claims

1. 1. A control system for controlling operation of an air conditioning system including a heat exchanger, comprising: at least one non-transitory memory configured to store computer-executable instructions; and at least one processor configured to execute the computer-executable instructions, the computer-executable instructions comprising: a neural network configured to execute a trained neural network to simulate operation of the heat exchanger in response to test control inputs and generate outputs of the simulation based on historical data defining a state of the heat exchanger, the historical data including a series of historical control inputs provided to the heat exchanger and a series of historical outputs of the operation of the heat exchanger corresponding to the series of historical control inputs, the neural network a first arm configured to process the test control input appended to the series of historical control inputs using a first combination of a convolutional network and a recurrent network trained to extract control features indicative of changes in the test control input from the series of historical control inputs; a second arm configured to process the series of historical control inputs paired with the series of historical outputs using a second combination of a convolutional network and a recurrent network trained to extract output features indicative of dynamic coupling between inputs and corresponding outputs of the operation of the heat exchanger; and a third arm configured to process the control characteristics and the output characteristics to predict a test output of the operation of the heat exchanger corresponding to the test control input, the computer-executable instructions further comprising: determining control commands to the air conditioning system based on the predicted test outputs of the simulation of the operation of the heat exchanger in response to the test control inputs; A control system configured to transmit the determined control commands to actuators of the air conditioning system.

2. 2. The control system of claim 1, wherein the first combination of a convolutional network and a recurrent network in the first arm and the second combination of a convolutional network and a recurrent network in the second arm of the neural network each include a bank of one-dimensional convolutional layers followed by a gated recurrent unit.

3. The control system of claim 1 , wherein the third arm is a fully connected deep neural network that accepts tensors including the extracted control features and the extracted output features.

4. 2. The control system of claim 1, wherein the determined control commands to the actuators of the air conditioning system are related to at least one of a compressor speed, a valve opening or closing, or a fan speed of the air conditioning system.

5. 2. The control system of claim 1, wherein a first neural weight is assigned to a first path of the first arm that processes the test control input appended to the series of historical control inputs, and a second neural weight is assigned to a second path of the second arm that processes the series of historical control inputs paired with the series of historical outputs.

6. The processor further comprises: generating one or more data sets associated with data trajectories obtained based on execution of a physics-based model of the heat exchanger having one or more boundary conditions associated with the heat exchanger; The control system of claim 1 , configured to apply a machine learning algorithm to the generated data set to train the neural network.

7. The processor further comprises:

7. The control system of claim 6, configured to generate a plurality of batches of trajectories of the data to generate the one or more data sets, each batch of the plurality of batches including at least one of the series of historical control inputs, the series of historical outputs, initial inputs to the air conditioning system, and true outputs of the air conditioning system.

8. The process of training the neural network comprises: calculating a mean square error for each batch of the plurality of batches based on the predicted output of the air conditioning system and the true output of the air conditioning system; determining a training loss associated with the mean squared error calculated for each batch of the plurality of batches; and optimizing the determined training loss.

9. The processor further comprises: and configured to initialize a model associated with the air conditioning system, the model including a room model and a plurality of component models of a plurality of components of the air conditioning system, the plurality of component models including at least the neural network trained to simulate the operation of the heat exchanger, the processor further comprising: determining one or more conditions associated with at least an air side of a space associated with the air conditioning system; executing the room model and the component models of the components of the air conditioning system based on the one or more conditions associated with at least an air side of the air conditioning system; 2. The control system of claim 1, configured to generate one or more states associated with the room model and each of the plurality of component models, the one or more states including at least one of a refrigerant state or an air state associated with the room model and each of the plurality of component models.

10. The processor further comprises: generating multiple test control inputs based on a Gaussian process; providing the plurality of test control inputs as inputs to a model of the air conditioning system; receiving a plurality of test outputs corresponding to the plurality of test control inputs predicted by the model of the air conditioning system; selecting the test control input from the generated plurality of test control inputs based on optimizing a cost function associated with the plurality of test control inputs and the plurality of test outputs; The control system of claim 1 , configured to determine the control commands to the air conditioning system based on the predicted test outputs of the simulation of the operation of the heat exchanger for selected test control inputs.

11. 11. The control system of claim 10, wherein the processor is further configured to utilize one or more test control inputs at a first time point to generate the plurality of test control inputs based on the Gaussian process for a second time point, the first time point being earlier than the second time point.

12. The processor further comprises: generating a set of test control inputs corresponding to a finite number of time points; providing the set of test control inputs as inputs to a model of the air conditioning system; receiving a set of test outputs corresponding to the set of test control inputs predicted by the model of the air conditioning system; calculating a gradient of a cost function associated with the set of test control inputs and the set of test outputs; selecting the test control input from the generated set of test control inputs based on the calculated gradient of the cost function; The control system of claim 1 , configured to determine the control commands to the air conditioning system based on the predicted test outputs of the simulation of the operation of the heat exchanger for selected test control inputs.

13. 1. A method for controlling operation of an air conditioning system including a heat exchanger, comprising: executing a neural network trained to simulate operation of the heat exchanger in response to test control inputs and to generate outputs of the simulation based on historical data defining a state of the heat exchanger, the historical data including a series of historical control inputs provided to the heat exchanger and a series of historical outputs of the operation of the heat exchanger corresponding to the series of historical control inputs, the neural network a first arm configured to process the test control input appended to the series of historical control inputs using a first combination of a convolutional network and a recurrent network trained to extract control features indicative of changes in the test control input from the series of historical control inputs; a second arm configured to process the series of historical control inputs paired with the series of historical outputs using a second combination of a convolutional network and a recurrent network trained to extract output features indicative of dynamic coupling between inputs and corresponding outputs of the operation of the heat exchanger; and a third arm configured to process the control characteristics and the output characteristics to predict a test output of the operation of the heat exchanger corresponding to the test control input, the method further comprising: determining control commands to the air conditioning system based on the predicted test outputs of the simulation of the operation of the heat exchanger in response to the test control inputs; and transmitting the determined control command to an actuator of the air conditioning system.

14. 14. The method of claim 13, wherein the first combination of a convolutional network and a recurrent network in the first arm and the second combination of a convolutional network and a recurrent network in the second arm of the neural network each include a bank of one-dimensional convolutional layers followed by a gated recurrent unit.

15. 14. The method of claim 13, wherein the third arm is a fully connected deep neural network that accepts tensors containing the extracted control features and the extracted output features.

16. generating one or more data sets associated with data trajectories obtained based on execution of a physics-based model having one or more boundary conditions associated with the heat exchanger; 14. The method of claim 13, further comprising applying a machine learning algorithm to the generated dataset to train the neural network.

17. 17. The method of claim 16, further comprising generating a plurality of batches of the data trajectories to generate the one or more data sets, each batch of the plurality of batches including at least one of the series of historical control inputs, the series of historical outputs, initial inputs to the air conditioning system, and true outputs of the air conditioning system.

18. and initializing a model associated with the air conditioning system, the model including a room model and a plurality of component models of a plurality of components of the air conditioning system, the plurality of component models including at least the neural network trained to simulate the operation of the heat exchanger; and determining one or more conditions associated with at least an air side of the air conditioning system; executing the room model and the component models of the components of the air conditioning system based on the one or more conditions associated with at least an air side of the air conditioning system; and generating one or more states associated with the room model and each of the plurality of component models, the one or more states including at least one of a refrigerant state or an air state associated with the room model and each of the plurality of component models.

19. The processor further comprises: generating multiple test control inputs based on a Gaussian process; providing the plurality of test control inputs as inputs to a model of the air conditioning system; receiving a plurality of test outputs corresponding to the plurality of test control inputs predicted by the model of the air conditioning system; selecting the test control input from the generated plurality of test control inputs based on optimizing a cost function associated with the plurality of test control inputs and the plurality of test outputs; 14. The method of claim 13, further comprising determining the control commands to the air conditioning system based on the predicted test outputs of the simulation of the operation of the heat exchanger for selected test control inputs.

20. A non-transitory computer-readable medium having stored thereon computer-executable instructions that, when executed by a computer, cause the computer to perform operations, the operations including: executing a neural network trained to simulate operation of a heat exchanger in response to test control inputs and to generate outputs of the simulation based on historical data defining a state of the heat exchanger, the historical data including a series of historical control inputs provided to the heat exchanger and a series of historical outputs of the operation of the heat exchanger corresponding to the series of historical control inputs, the neural network a first arm configured to process the test control input appended to the series of historical control inputs using a first combination of a convolutional network and a recurrent network trained to extract control features indicative of changes in the test control input from the series of historical control inputs; a second arm configured to process the series of historical control inputs paired with the series of historical outputs using a second combination of a convolutional network and a recurrent network trained to extract output features indicative of dynamic coupling between inputs and corresponding outputs of the operation of the heat exchanger; and a third arm configured to process the control characteristics and the output characteristics to predict a test output of the operation of the heat exchanger corresponding to the test control input, the operation further comprising: determining control commands to an air conditioning system based on the predicted test outputs of the simulation of the operation of the heat exchanger in response to the test control inputs; and transmitting the determined control command to an actuator of the air conditioning system.

Images