Implementation of a machine learning based surrogate model for indoor air flow and temperature prediction

Abstract

A surrogate model for predicting computational fluid dynamics simulations uses a computational fluids dynamics (CFD) simulator and a feedforward neural network. CFD simulator receives first input parameters of ambient conditions and operating conditions for an indoor space and generates heat maps and airflow maps of the indoor space. Target region data and an image background map are extracted from the seed images. Feedforward neural network is trained as a surrogate machine learning (ML) based model to learn normalized CFD prediction data based on the first input parameters. Training is based on a loss function comparison of the normalized CFD prediction data to the normalized target region data. The surrogate model generates normalized CFD prediction data from received second input parameters of ambient conditions and operating conditions for the indoor space. CFD prediction images are generated by denormalizing the prediction data and combining with the image background map.

Description

IMPLEMENTATION OF A MACHINE LEARNING BASED SURROGATE MODEL FOR INDOOR AIR FLOW AND TEMPERATURE PREDICTIONTECHNICAL FIELD

[0001] This application relates to neural networks. More particularly, this application relates to applying a machine learning framework for simulating computational fluid dynamics related to indoor air flow and temperature prediction.BACKGROUND

[0002] Existing approaches for designing complex devices and systems in building automation, car manufacturing, and public transportation involve an iterative interaction between exploration of the design / operating space and evaluation of the performance using high-fidelity Computational Fluid Dynamics (CFD) simulations. For example, designing an HVAC system for rooms of a building or transportation compartments involves computationally expensive CFD simulation of temperature and airflow inside the building under a variety of design, ambient, and operating conditions. However, the computational cost associated with such high-fidelity CFD simulations severely limits the scope of the overall design optimization process. As a consequence, it can lead to sub- optimal design choices.

[0003] Traditionally, design approaches use high-fidelity commercial CFD solvers, for example, SimCenter Star-CCM+ or Ansys Fluent, to predict indoor temperature for a given design (e.g., geometry of a car cabin), ambient condition (e.g., outdoor temperature, solar load and orientation) and operating condition (e.g., HVAC settings). However, suchCFD simulations are computationally very expensive and need significant computing time and power.SUMMARY

[0004] A surrogate model for predicting computational fluid dynamics simulations uses a computational fluids dynamics (CFD) simulator and a feedforward neural network. CFD simulator receives first input parameters of ambient conditions and operating conditions for an indoor space and generates heat maps and airflow maps of the indoor space. Target region data and an image background map are extracted from the seed images. Feedforward neural network is trained as a surrogate machine learning (ML) based model to learn normalized CFD prediction data based on the first input parameters. Training is based on a loss function comparison of the normalized CFD prediction data to the normalized target region data. The surrogate model generates normalized CFD prediction data from received second input parameters of ambient conditions and operating conditions for the indoor space. CFD prediction images are generated by denormalizing the prediction data and combining with the image background map.BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Non-limiting and non-exhaustive embodiments of the present disclosure are described with reference to the following FIGURES, wherein like reference numerals refer to like elements throughout the drawings unless otherwise specified.

[0006] FIG. 1 illustrates a framework for a ML-based surrogate model used for prediction of indoor air flow and temperature in accordance with embodiments of this disclosure.

[0007] FIG. 2 illustrates an example of a process for training and operating the surrogate model depicted in FIG. 1 in accordance with embodiments of this disclosure.

[0008] FIG. 3 shows an example of a series of seed simulations in accordance with embodiments of this disclosure.

[0009] FIG. 4 shows an example of a computing environment in which embodiments of the present disclosure may operate.DETAILED DESCRIPTION

[0010] Methods and systems disclosed herein use neural networks to approximate functional relationships and apply machine learning (ML) algorithms to learn a fast and accurate surrogate model that can predict CFD simulation results. By providing a means to accelerate the design evaluations significantly, the disclosed framework improves design choices that remain performant over a wide range of ambient and operating conditions. Furthermore, the disclosed framework has potential application over a broad range of domains (e.g., automotive, public transport, smart buildings). A technical advantage provided by the proposed approach over existing solutions lies in the way temperature distribution is predicted. Traditional approaches use an iterative numerical solver to simulate the indoor airflow for predicting temperature and air speed. In contrast, the proposed approach predicts indoor temperature and air speed through a neural network inference which involves only a single forward pass through the neural network and can be done in a near real-time manner.

[0011] Another technical advantage of the proposed approach is conservation of computing resources. To explore the relevant design space of an HVAC system for abuilding using traditional approaches need to run computationally expensive and time consuming CFD simulation for all relevant combinations of input parameter values. This significantly increases the time and cost associated with design exploration. In contrast, the disclosed approach uses only few values of the input parameter values to create a “seed” dataset which is then utilized to train a neural network based surrogate model. The trained surrogate model is then used to simulate the indoor airflow for any combination of the input parameter values.

[0012] FIG. 1 illustrates an example of a framework for a ML-based surrogate model to predict indoor air flow and temperature in accordance with embodiments of this disclosure. FIG. 2 illustrates an example of a method for training and operating the surrogate model depicted in FIG. 1. As an illustrative example, the framework 100 is described in terms of generating two dimensional (2D) images (e.g., heat maps or airflow maps) representing predicted temperature or airflow in an indoor environment (e.g., building, automobile, aircraft).

[0013] In a first step 210, framework 100 identifies the input parameters 105, e.g., ambient conditions (e.g., solar load, sun altitude, sun azimuth, and outside temperature) and operating (e.g., discharge air temperature, volume flow rate, etc.) conditions and the ranges over which their values can vary. At 220, multiple values of these parameters are sampled from the relevant range and combinations are used to generate “seed” simulation results using CFD simulator 110 (e.g., Simcenter StarCCM+). The format for the seed results (e.g., temperature, air speed, etc.) on relevant cross-sectional planes are 2D images 1 15. FIG. 3 shows an example of a series of seed simulations results in theform of a 2D heat map for an automobile cabin showing eight prediction model maps and one ground truth map.

[0014] At 230, a preprocessing step is applied to these seed simulation results which have been saved as 2D images 1 15. In particular, target region extractor 120 removes the image background to identify the target region. The image background is saved as background map 122. Next at 240, data normalizer 124 operates on the target region over the whole dataset used for training. Data from 2D images are translated by a scaling operation to generate data values from the scalar image representation (e.g., translate a heat map to an array of temperature values). The scaling operation removes the mean value from all the data points, then scales the values by the inverse of standard deviation, giving normalized values with mean=0 and variance=1 .

[0015] At 250, a feedforward neural network (NN) 125 with (L hidden layers and N neurons / layer) is trained to predict normalized data values over the target region. The aforementioned ambient and operating conditions constitute the input to this neural network, and the output of this neural network is the predicted distribution of normalized data (i.e., normalized temperature, normalized air speed, etc.) over the target region.

[0016] Neural network 125 is trained by employing the following algorithm choices: (a) a Loss Function 127 (e.g., L1 or L2 loss function) uses the mean absolute error (MAE) between the normalized predicted values 126 and true values from data normalizer 124 to update the neural network parameters through backpropagation; (b) dropout layers improve the generalization performance of the learned ML-based surrogate and preventing overfitting by randomly setting input units to 0 with a user-defined probability (the "dropout rate"); the sum over all inputs are kept unchanged by scaling the inputs thatare not set to 0 by 1 / (1 - dropout rate) and (c) a learning rate scheduling algorithm (e.g., cosine Annealing) is used to achieve fast and convergent training behavior.

[0017] During training phase, different ML-based models are trained on the data set with respect to L layers and N neurons / layer as well as algorithm choices [(a), (b), (c)] and the model with the best performance (i.e., highest accuracy over the test data set) is selected to finalize the network architecture using grid-search and cross-validation. As an example of cross validation, the training data set may be divided into k subsets, using one subset as a test set and k-1 subsets as training data. The grid search algorithm attempts all possible combinations of NN parameter values and returns the combination with the highest accuracy.

[0018] For the design phase, at 260, the trained NN 131 is fed input parameters 135 to synthesize predicted 2D images 140 of relevant simulation variable(s) (e.g., temperature, air speed, etc.) on the cross-sectional plane. Normalized predictions 136 are denormalized by denormalizer 137 by scaling the predicted outputs by the standard deviation and then adds the mean value. This denormalized data is fed to image generator 138 which adds the image background using background map 122 to yield the 2D images 140 of the desired simulation variable (e.g., heat map, airflow map) on a cross sectional plane.

[0019] The main difference between the proposed approach and existing solutions lies in the way they predict the temperature distribution. Traditional approaches use an iterative numerical solver to simulate the indoor airflow for predicting temperature and air speed. On the other hand, the proposed approach predicts indoor temperature and airspeed through a neural network inference (which involves only a single forward pass through the neural network and can be done in a near real-time manner).

[0020] To explore the relevant design space of an HVAC system for a building using traditional approaches need to run computationally expensive and time consuming CFD simulation for all relevant combinations of input parameter values. This significantly increases the time and cost associated with design exploration. In contrast, the approach by the disclosed embodiments uses only few values of the input parameter values to create a “seed” dataset which is then utilized to train a neural network based surrogate model. The trained surrogate model is then used to simulate the indoor airflow for any combination of the input parameter values.

[0021] FIG. 4 shows an example of a computing environment in which embodiments of the present disclosure may operate. Computer system 410 may be embodied, for example and without limitation, as a computing device for synthesizing controllers of underactuated systems.

[0022] Processors 415 may include one or more GPUs and one or more central processing units (CPUs). System memory 416 stores information and instructions to be executed by processors 415 and may be used for storing temporary variables or other intermediate information during the execution of instructions by processors 415. System memory 416 may contain data and / or program modules that are immediately accessible to and / or presently being operated on by the processors 415, such as CFD Simulator module 41 1 , neural network module 412, and image processing module 413. For this example, module 411 is configured to execute the functionality of the CFD simulator 1 10 and NN module 412 is configured to execute the functionality of the neural network 125,131 as described above with reference to FIG. 1. Image processing module 413 is configured with the functionality of target region extractor 120, data normalizer 124, denormalizer 137 and image generator 138 as described above with reference to FIG. 1 .

[0023] Computing system 410 may also include a user interface module 423 for communicating with a graphical user interface 424 that includes a display device to display information to a computer user, one or more input devices, such as a keyboard or pointing device, for interacting with a computer user and providing information to the processors 415. The display device may provide a touch screen interface which allows input to supplement or replace the communication of direction information and command selections.

[0024] The computing system 410 may perform a portion or all of the processing steps of embodiments of the disclosure in response to the processors 415 executing one or more sequences of one or more instructions contained in a memory, such as the system memory 416. Such instructions may be read into the system memory 416 from another computer readable storage medium, such as local storage device 422, implemented as a magnetic hard disk or a removable media drive. The local storage device 422 may contain one or more datastores and data files used by embodiments of the present disclosure. Datastore contents and data files may be encrypted to improve security. The processors 415 may also be employed in a multi-processing arrangement to execute the one or more sequences of instructions contained in system memory 416. In alternative embodiments, hard-wired circuitry may be used in place of or in combination with software instructions. Thus, embodiments are not limited to any specific combination of hardwarecircuitry and software. Additionally, or alternatively, data storage, such as training data, may be stored and accessed from remote location 431 via network 450.

[0025] The computing system 410 may include at least one computer readable storage medium or memory, such as local storage device 422, for holding instructions programmed according to embodiments of the disclosure and for containing data structures, tables, records, or other data described herein. The term “computer readable storage medium” as used herein refers to any medium that participates in providing instructions to the processor 415 for execution. A computer readable storage medium may take many forms including, but not limited to, non-transitory, non-volatile media, volatile media, and transmission media. Non-limiting examples of non-volatile media include optical disks, solid state drives, magnetic disks, and magneto-optical disks, such as magnetic hard disk or removable media drive. Non-limiting examples of volatile media include dynamic memory, such as system memory 416. Non-limiting examples of transmission media include coaxial cables, copper wire, and fiber optics. Transmission media may also take the form of acoustic or light waves, such as those generated during radio wave and infrared data communications. Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network.

[0026] Network 450 may be any network or system generally known in the art, including the Internet, an intranet, a local area network (LAN), a wide area network (WAN), a metropolitan area network (MAN), a direct connection or series of connections,a cellular telephone network, or any other network or medium capable of facilitating communication between computing system 410 and other computers such as remote computing device 441 . The network 450 may be wired, wireless or a combination thereof. Wired connections may be implemented using Ethernet, Universal Serial Bus (USB), RJ- 6, or any other wired connection generally known in the art. Wireless connections may be implemented using Wi-Fi, WiMAX, and Bluetooth, infrared, cellular networks, satellite or any other wireless connection methodology generally known in the art. Additionally, several networks may work alone or in communication with each other to facilitate communication in the network 450.

[0027] The embodiments of the present disclosure may be implemented with any combination of hardware and software. In addition, the embodiments of the present disclosure may be included in an article of manufacture (e.g., one or more computer program products) having, for example, a non-transitory computer-readable storage medium. The computer readable storage medium has embodied therein, for instance, computer readable program instructions for providing and facilitating the mechanisms of the embodiments of the present disclosure. The article of manufacture can be included as part of a computer system or sold separately.

[0028] Computer readable medium instructions for carrying out operations of the present disclosure may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like, and conventional proceduralprogramming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions may execute entirely on the computing device, partly on the computing device, as a stand-alone software package, partly on the computing device and partly on a remote computer or entirely on the computing device or server. In the latter scenario, the remote computer may be connected to the computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) may execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present disclosure.

[0029] Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, may be implemented by computer readable medium instructions.

[0030] The program modules, applications, computer-executable instructions, code, or the like depicted in FIG. 4 as being stored in the system memory 416 are merely illustrative and not exhaustive and that processing described as being supported by any particular module may alternatively be distributed across multiple modules or performedby a different module. In addition, various program module(s), script(s), plug-in(s), Application Programming Interface(s) (API(s)), or any other suitable computer-executable code hosted locally on the computing system 410, and / or hosted on other computing device(s) accessible via one or more of network, may be provided to support functionality provided by the program modules, applications, or computer-executable code and / or additional or alternate functionality. Further, functionality may be modularized differently such that processing described as being supported collectively by the collection of program modules 41 1 , 412, 413 may be performed by a fewer or greater number of modules, or functionality described as being supported by any particular module may be supported, at least in part, by another module. In addition, program modules that support the functionality described herein may form part of one or more applications executable across any number of systems or devices in accordance with any suitable computing model such as, for example, a client-server model, a peer-to-peer model, and so forth. In addition, any of the functionality described as being supported by any of the program modules depicted in FIG. 4 may be implemented, at least partially, in hardware and / or firmware across any number of devices.

[0031] It should further be appreciated that the computing system 410 may include alternate and / or additional hardware, software, or firmware components beyond those described or depicted without departing from the scope of the disclosure. More particularly, it should be appreciated that software, firmware, or hardware components depicted as forming part of the computing system 410 are merely illustrative and that some components may not be present or additional components may be provided in various embodiments. While various illustrative program modules have been depictedand described as software modules stored in system memory 416, it should be appreciated that functionality described as being supported by the program modules may be enabled by any combination of hardware, software, and / or firmware. It should further be appreciated that each of the above-mentioned modules may, in various embodiments, represent a logical partitioning of supported functionality. This logical partitioning is depicted for ease of explanation of the functionality and may not be representative of the structure of software, hardware, and / or firmware for implementing the functionality. Accordingly, it should be appreciated that functionality described as being provided by a particular module may, in various embodiments, be provided at least in part by one or more other modules. Further, one or more depicted modules may not be present in certain embodiments, while in other embodiments, additional modules not depicted may be present and may support at least a portion of the described functionality and / or additional functionality. Moreover, while certain modules may be depicted and described as submodules of another module, in certain embodiments, such modules may be provided as independent modules or as sub-modules of other modules.

[0032] Although specific embodiments of the disclosure have been described, one of ordinary skill in the art will recognize that numerous other modifications and alternative embodiments are within the scope of the disclosure. For example, any of the functionality and / or processing capabilities described with respect to a particular device or component may be performed by any other device or component. Further, while various illustrative implementations and architectures have been described in accordance with embodiments of the disclosure, one of ordinary skill in the art will appreciate that numerous other modifications to the illustrative implementations and architecturesdescribed herein are also within the scope of this disclosure. In addition, it should be appreciated that any operation, element, component, data, or the like described herein as being based on another operation, element, component, data, or the like can be additionally based on one or more other operations, elements, components, data, or the like. Accordingly, the phrase “based on,” or variants thereof, should be interpreted as “based at least in part on.”

[0033] The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.

Claims

CLAIMSWhat is claimed is:1 . A computer-implemented system for creating a surrogate model for predicting computational fluid dynamics simulations, the system comprising: a processor; and memory having modules stored thereon with instructions to be executed by the processor, the modules comprising: a computational fluids dynamics (CFD) simulator configured to: receive first input parameters of ambient conditions and operating conditions for an indoor space; generate seed simulation images representing heat maps of the indoor space and second seed simulation images representing airflow maps of the indoor space; an image processing module configured to: extract target region data and an image background map from the seed images; normalize the target region data; a feedforward neural network trained as a surrogate machine learning (ML) based model to learn normalized CFD prediction data based on received input parameters of ambient and operating conditions of the indoor space, wherein the training is based on a loss function comparison of the normalized CFD prediction data to the normalized target region data, the trained feedforward neural network configured to: generate normalized CFD prediction data from received second input parameters of ambient conditions and operating conditions for the indoor space;the image processing module further configured to: denormalize the normalized CFD prediction data; generate 2D images of CFD predict data by combining the image background map to the denormalized data.

2. The system of claim 1 , wherein a final surrogate model is selected from different ML based models trained on the dataset with respect to L layers and N neurons per layer based on best performance over a test data set, the selection based on grid-search and cross-validation.

3. The system of claim 1 , wherein the training comprises the loss function using mean absolute error (MAE) between the normalized CFD prediction data and the normalized target region data to update the neural network parameters through backpropagation.

4. The system of claim 1 , wherein the training comprises: using dropout layers to improve the generalization performance of surrogate model.

5. The system of claim 1 , wherein the training comprises: a learning rate scheduling algorithm to achieve fast and convergent training behavior.

6. A computer-implemented method for creating a surrogate model for predicting computational fluid dynamics simulations, the method comprising: receiving first input parameters of ambient conditions and operating conditions for an indoor space;generating seed simulation images representing heat maps of the indoor space and second seed simulation images representing airflow maps of the indoor space; extracting target region data and an image background map from the seed images; normalizing the target region data; training a feedforward neural network as a surrogate machine learning (ML) based model to learn normalized CFD prediction data based on received input parameters of ambient and operating conditions of the indoor space, wherein the training is based on a loss function comparison of the normalized CFD prediction data to the normalized target region data generating, by the trained feedforward neural network, normalized CFD prediction data from received second input parameters of ambient conditions and operating conditions for the indoor space; denormalizing the normalized CFD prediction data; generating 2D images of CFD predict data by combining the image background map to the denormalized data.

7. The method of claim 6, further comprising: selecting a final surrogate model from different ML based models trained on the dataset with respect to L layers and N neurons per layer based on best performance over a test data set, the selection based on grid-search and cross- validation.

8. The method of claim 6, wherein the training comprises the loss function using mean absolute error (MAE) between the normalized CFD prediction data and the normalized target region data to update the neural network parameters through backpropagation.

9. The method of claim 6, wherein the training comprises: using dropout layers to improve the generalization performance of surrogate model.

10. The method of claim 6, wherein the training comprises: a learning rate scheduling algorithm to achieve fast and convergent training behavior.

Images