News | Application Package Repository Telkom University

Category: News

Bookings Link not displaying correct times

/ 2024-05-16

I have been experiencing this issue intermittently for the past 6 months or so. In my booking page, it shows that I have 6 separate slots, but when I send the link to somebody else, it only shows 2 of the 6 times as available.

Changing a Drop Down component items via a string.

/ 2024-05-16 / Matlab

Hi all,
I am making a GUI for my XML files. The problem is related to a Drop Down component. As shown in the picture:

In the last column (option) and second row, a 1×6 string exist. Here is ther problem. When I manually define a 1×6 string array using the code bellow:
B = ["Red","Yellow","Blue","Green"];
I can change component’s items. But, when I use the bellow opcode:
inStruct.input.group.group(1).param(2).option
An error emerges!
Anyway, my question is how it is possible to change my drop down menu items by the items included in the second row and last column?

TNX.Hi all,
I am making a GUI for my XML files. The problem is related to a Drop Down component. As shown in the picture:

TNX. Hi all,
I am making a GUI for my XML files. The problem is related to a Drop Down component. As shown in the picture:

TNX. xml, gui MATLAB Answers — New Questions

Error in building C++ shared Library

/ 2024-05-16 / Matlab

I am getting this error, when I am building cpp shared library with dlls.

Building with ‘Microsoft Visual C++ 2017’
LINK: fatal error LNK1181: cannot open inputt file ‘gdi32.lib’

Error using compiler.internal.build.builder.Generic/build
The following command:
""C:Program FilesMATLABR2023abinmbuild" -client mbuild -O -output "mwa_Addmatrix" -I".mwa_AddmatrixcppSharedLibrary"
".mwa_AddmatrixcppSharedLibrarymwa_Addmatrix.cpp" DEFFILE"="".mwa_AddmatrixcppSharedLibrarymwa_Addmatrix.def""" LINKEXPORT"=""/DLL
/def:"$DEFFILE"""" LDEXT"="".dll""" LDTYPE"="" -shared -static""" MTCMDLINE"=""mt -outputresource:$EXE;2 -manifest $MANIFEST""" -outdir
".mwa_AddmatrixcppSharedLibrary""
failed with error code -1.

Error in compiler.build.cppSharedLibrary (line 115)
resultsobj = builder.build;I am getting this error, when I am building cpp shared library with dlls.

Building with ‘Microsoft Visual C++ 2017’
LINK: fatal error LNK1181: cannot open inputt file ‘gdi32.lib’

Error in compiler.build.cppSharedLibrary (line 115)
resultsobj = builder.build; I am getting this error, when I am building cpp shared library with dlls.

Building with ‘Microsoft Visual C++ 2017’
LINK: fatal error LNK1181: cannot open inputt file ‘gdi32.lib’

Error in compiler.build.cppSharedLibrary (line 115)
resultsobj = builder.build; cpp shared library, cpp, matlab, dll, compiler, sharedlibrary, library compiler, c++ MATLAB Answers — New Questions

Library compiler Error while packaging MATLAB functions into a C++ library and create a Dynamic Linking Library (DLL)

/ 2024-05-16 / Matlab

I am converting MATLAB functions into a C++ library and create a Dynamic Linking Library (DLL) using LIbrary Compiler. I am using Matlab 2023a and Microsoft Visual Studio 2017. It was working fine. But when I formatted the system and reinstalled everything, it is giving the following error while packaging-
""C:Program FilesMATLABR2023abinmbuild" -client mbuild -O -v -output "test" -I"D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testing" "D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testingtest.cpp" DEFFILE"=""D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testingtest.def""" LINKEXPORT"=""/DLL /def:"$DEFFILE"""" LDEXT"="".dll""" LDTYPE"="" -shared -static""" MTCMDLINE"=""mt -outputresource:$EXE;2 -manifest $MANIFEST""" -outdir "D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testing""
failed with error code -1.

mcc failed

Please suggest, where am I going wrong.I am converting MATLAB functions into a C++ library and create a Dynamic Linking Library (DLL) using LIbrary Compiler. I am using Matlab 2023a and Microsoft Visual Studio 2017. It was working fine. But when I formatted the system and reinstalled everything, it is giving the following error while packaging-
""C:Program FilesMATLABR2023abinmbuild" -client mbuild -O -v -output "test" -I"D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testing" "D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testingtest.cpp" DEFFILE"=""D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testingtest.def""" LINKEXPORT"=""/DLL /def:"$DEFFILE"""" LDEXT"="".dll""" LDTYPE"="" -shared -static""" MTCMDLINE"=""mt -outputresource:$EXE;2 -manifest $MANIFEST""" -outdir "D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testing""
failed with error code -1.

mcc failed

Please suggest, where am I going wrong. I am converting MATLAB functions into a C++ library and create a Dynamic Linking Library (DLL) using LIbrary Compiler. I am using Matlab 2023a and Microsoft Visual Studio 2017. It was working fine. But when I formatted the system and reinstalled everything, it is giving the following error while packaging-
""C:Program FilesMATLABR2023abinmbuild" -client mbuild -O -v -output "test" -I"D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testing" "D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testingtest.cpp" DEFFILE"=""D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testingtest.def""" LINKEXPORT"=""/DLL /def:"$DEFFILE"""" LDEXT"="".dll""" LDTYPE"="" -shared -static""" MTCMDLINE"=""mt -outputresource:$EXE;2 -manifest $MANIFEST""" -outdir "D:SCATTER_CORRECTIONkernel_basedsc_dll_thoraxtestfor_testing""
failed with error code -1.

mcc failed

Please suggest, where am I going wrong. library compiler, dll, c++ shared library, matlab compiler, embedded matlab function, compiler, application compiler, cpp MATLAB Answers — New Questions

Installing standalone .exe application using matlab runtime installer.

/ 2024-05-16 / Matlab

I have packaged a standalone application using Matlab Compiler(R2020a). At that time, I did not have option to choose the matlab runtime version with which I want to package my application. Is it possible to install an app packaged with matlab runtime 9.8 in another device which has matlab runtime 9.4? if not is there any other way to do it??I have packaged a standalone application using Matlab Compiler(R2020a). At that time, I did not have option to choose the matlab runtime version with which I want to package my application. Is it possible to install an app packaged with matlab runtime 9.8 in another device which has matlab runtime 9.4? if not is there any other way to do it?? I have packaged a standalone application using Matlab Compiler(R2020a). At that time, I did not have option to choose the matlab runtime version with which I want to package my application. Is it possible to install an app packaged with matlab runtime 9.8 in another device which has matlab runtime 9.4? if not is there any other way to do it?? matlab compiler, matlab gui MATLAB Answers — New Questions

Need help on an error regarding a Simulink model

/ 2024-05-16 / Matlab

I have a ongoing study on a model regarding insulation coordination which I downloaded from this link https://www.mathworks.com/matlabcentral/fileexchange/108549-insulation-coordination-in-a-generating-station/?s_tid=LandingPageTabfx, it shows an error when I tried to run the model. Not sure what I’m doing wrong, hoping for a solution.
I have attached the original file(project1F30X2_m) and a slightly modified version(model2).
Error Encountered: Specialized Power Systems cannot solve this circuit. Please check for one the following two possibilities : 1) The circuit solution is undetermined. This problem arises when transformers with no magnetization branch (Rm=inf; Lm=inf) are connected together. –> Specify a magnetizing branch with finite Rm and/or Lm parameter values. 2) Circuit solution is inuccurate due to a badly scaled circuit. This is usually caused by a too wide range of resistance values (ex 1e-6 ohm and 1e6 ohm in the same circuit). –> Try to reduce range of resistance values. For example, reduce snubber resistances.I have a ongoing study on a model regarding insulation coordination which I downloaded from this link https://www.mathworks.com/matlabcentral/fileexchange/108549-insulation-coordination-in-a-generating-station/?s_tid=LandingPageTabfx, it shows an error when I tried to run the model. Not sure what I’m doing wrong, hoping for a solution.
I have attached the original file(project1F30X2_m) and a slightly modified version(model2).
Error Encountered: Specialized Power Systems cannot solve this circuit. Please check for one the following two possibilities : 1) The circuit solution is undetermined. This problem arises when transformers with no magnetization branch (Rm=inf; Lm=inf) are connected together. –> Specify a magnetizing branch with finite Rm and/or Lm parameter values. 2) Circuit solution is inuccurate due to a badly scaled circuit. This is usually caused by a too wide range of resistance values (ex 1e-6 ohm and 1e6 ohm in the same circuit). –> Try to reduce range of resistance values. For example, reduce snubber resistances. I have a ongoing study on a model regarding insulation coordination which I downloaded from this link https://www.mathworks.com/matlabcentral/fileexchange/108549-insulation-coordination-in-a-generating-station/?s_tid=LandingPageTabfx, it shows an error when I tried to run the model. Not sure what I’m doing wrong, hoping for a solution.
I have attached the original file(project1F30X2_m) and a slightly modified version(model2).
Error Encountered: Specialized Power Systems cannot solve this circuit. Please check for one the following two possibilities : 1) The circuit solution is undetermined. This problem arises when transformers with no magnetization branch (Rm=inf; Lm=inf) are connected together. –> Specify a magnetizing branch with finite Rm and/or Lm parameter values. 2) Circuit solution is inuccurate due to a badly scaled circuit. This is usually caused by a too wide range of resistance values (ex 1e-6 ohm and 1e6 ohm in the same circuit). –> Try to reduce range of resistance values. For example, reduce snubber resistances. error MATLAB Answers — New Questions

Microsoft

Dax measure not returning the correct aggregate result

/ 2024-05-16

Hi @Sergei Baklan

I created this measure shown below that returns a table with correct row values:

But the aggregate total -3036.53 is not correct and correct total we need is -1145.19.

Could you please help me as what needs to be corrected in the above measure for the correct total?

PFA file here Portfolio Performance – v2.12 – Copy (2).pbix

Thanks in advance sir!

Hi @Sergei Baklan I created this measure shown below that returns a table with correct row values: But the aggregate total -3036.53 is not correct and correct total we need is -1145.19.Could you please help me as what needs to be corrected in the above measure for the correct total? PFA file here Portfolio Performance – v2.12 – Copy (2).pbix Thanks in advance sir! Read More

I am trying to stop all my scheduled triggers in dev.

/ 2024-05-16 / Microsoft

Folks, I am trying to stop all my scheduled triggers for ADF pipeline run in dev. Even if the trigger status is stopped, it runs every day as per the schedule. Any idea how to stop them.

Folks, I am trying to stop all my scheduled triggers for ADF pipeline run in dev. Even if the trigger status is stopped, it runs every day as per the schedule. Any idea how to stop them. Read More

Default Subject line when sending invitations from Personal Bookings page

/ 2024-05-16 / Microsoft

Hey,

When I use the link on my personal Bookings page associated with my work email to send a private invitation, the default subject line appears to be “Book time with me without back-and-forth discussions.” Is there any way to change that? I’m unable to find an edit option. I do have admin rights to our tenant if I need them.

Hey, When I use the link on my personal Bookings page associated with my work email to send a private invitation, the default subject line appears to be “Book time with me without back-and-forth discussions.” Is there any way to change that? I’m unable to find an edit option. I do have admin rights to our tenant if I need them. Read More

Microsoft

🦸🏾‍♀️Analyze your communications: Superhero to Organization Leader Analysis Prompt evolution

/ 2024-05-16

HLS Copilot Snacks are the perfect way to learn how to use Copilot to boost your productivity and reduce your workload.

In this snackable you will learn how you can analyze your emails, meetings, and chats to understand who your superhero, anime character, and leadership style is. This video goes through the evolution of prompting on a fun topic to one that is focused on professional growth.

To see all HLS Copilot Snacks video click here.

Prompts Used within Word

🦸🏾‍:female_sign:Superhero

Look at the emails and Teams meetings that I have sent in the last two weeks. Use these messages to distill a personal brand voice document that I can use to inform copilot of my personal voice and style. Based on the above analysis, what superhero represents my style of work best? What would my super powers be? Can you format that as a dossier for superheroes — listing my real name, and most similar existing superhero, and include the rest of the information formatted and edited in dossier style?

Anime Character

Look at the emails and Teams meetings that I have sent in the last two weeks. Use these messages to distill a personal brand voice document that I can use to inform copilot of my personal voice and style based on the above analysis, what (Genre: example Magical pretty girl) Anime Character would represents my style of work best and why? What would my abilities and personality flaws be? Listing my real name and most similar existing Anime Character. Also give me a anime character friend group, they should be from different anime, include a table with |Name | Anime| strengths | flaws | why they are my friends | how they help me or make my flaws worst. Also be sure to identify my rival anime character and a short story of how we became enemies but may end up being friends.

Leadership

Style, Strengths, and Goals Look at the emails, teams meetings, documents, teams chats that I’ve sent in the last year. Use these messages to distill a leadership style document that I can use to inform copilot of my personal voice and style based on the above analysis. Give me a breakdown of my top 5 strengths in a Table with the following formation | Strength | description | how it helps me in my role | how it could be misunderstood | 2 ways I can build this strength. In another table give me a breakdown of 3 areas for improvement in the following formation | Improvement area | description | how it impacts my role | how it could be misunderstood | 3 ways I can improve. I would also like 3 OKRs based on the above analysis that I could reach in the next 6-9 months.

Have questions you would like to have us address in a snackable? Let us know! 

Thanks for Visiting – Jasmine Hoegh, PMP, Connect with me on LinkedIn

Microsoft Tech Community – Latest Blogs –Read More

Microsoft

Loop DDoS Attacks: Understanding the Threat and Azure’s Defense

/ 2024-05-16

In the realm of cybersecurity, Distributed Denial-of-Service (DDoS) attacks are a significant concern. The recent holiday season has unveiled a complex and evolving threat landscape, marked by sophisticated tactics and diversification. From botnet delivery via misconfigured Docker API endpoints to the NKAbuse malware’s exploitation of blockchain technology for DDoS attacks, the tactics and scale of these attacks have shown significant sophistication and diversification.

Understanding and staying abreast of recent DDoS trends and attack vectors is crucial for maintaining robust network security and ensuring the availability of services. One such example is the recent HTTP/2 Rapid Reset Attack, where Microsoft promptly provided fixes and recommendations to safeguard web applications. This vulnerability exploits the HTTP/2 protocol, allowing attackers to disrupt server connections by rapidly opening and closing connection streams. This can lead to denial of service (DoS) conditions, severely impacting the availability of critical services and potentially leading to significant downtime and financial losses. Another example we wrote about were reflected TCP attack vectors that recently emerged in ways that were not believed possible before.

By closely monitoring these emerging threats, security professionals can develop and implement timely and effective countermeasures to protect their networks. This proactive approach is essential for anticipating potential vulnerabilities and mitigating risks before they can be exploited by malicious actors. Furthermore, understanding the evolving landscape of DDoS attacks enables the development of more resilient security architectures and the enhancement of existing defense mechanisms, ensuring that networks remain secure against both current and future threats.

In this blog, we focus on the newly revealed Application Loop DDoS attack vector. Microsoft hasn’t witnessed this vulnerability translated to actual DDoS attacks yet. However, we believe it’s important to highlight the threat landscape we see in Azure for UDP reflected attacks, as they present a prevalent attack vector with similar base pattern as Loop attacks. We then discuss what protection strategies Microsoft employs to protect Azure platform, our online services, and customers from newly emerging threats.

The Emergence of Loop DDoS Attacks

The Loop attack vulnerability was disclosed last month by CISPA. The attack exploits application-layer protocols relying on User Datagram Protocol (UDP). CISPA researchers found ~300,000 application servers that may be vulnerable to this attack vector. The published advisory describes Loop attacks as a sophisticated DDoS vector, exploiting the interaction between application servers to create a never-ending (hence the term Loop) cycle of communication that can severely degrade or completely halt their functionality. This attack method uses spoofed attack sources to create a situation where two or more application servers get stuck in a continuous loop of messages, usually error responses, because each server is programmed to react to incoming error messages with an error message.

Amongst the vulnerable applications, TFTP, DNS, NTP as well as legacy protocols, such as Echo, Chargen, QOTD, are at risk. The researchers provided a practical example of this, when two DNS resolvers automatically reply to error messages with their own errors. An attacker can start a loop by sending one fake spoofed DNS error to one resolver. This makes it send an error to the spoofed resolver, which does the same, creating an endless cycle of errors between them. This wastes the DNS servers’ resources and fills up the network links between them, with the potential to cause serious problems in service and network quality. Depending on the exact attack topology, Loop attacks may generate excessive amounts of traffic like other volumetric DDoS floods (e.g. DNS reflected amplified attacks).

How Loop DDoS differs from other volumetric DDoS attacks

The Loop attack is a kind of DDoS attack vector that targets applications and may manifest as a large-scale flood at the network layer as well. The cause is that attackers can set up multiple attack loops among multiple servers in a network or across networks in the peering links, overwhelming the servers and networks with traffic floods.

Like UDP reflected attacks, Loop attacks use a basic UDP weakness – the possibility to fake a source IP address to initiate the attack Loop. One of the most common attack vectors nowadays is the reflected UDP-based floods. It’s similar to Loop attack in that the malicious actor sends spoofed-source packets to an application server that replies to the spoofed IP, i.e. the victim. By generating many of these requests to an application server, the victim gets many of the responses they didn’t ask for. The impact of the reflected attack may be significantly more disastrous if the attacked application generates more traffic in response that it receives in the request. When this happens, it becomes a reflected amplified attack. Amplification is the secret sauce of why these attacks are dangerous. Loop attack is different than reflected amplified attacks in that the response may not necessarily be amplified. That is, for each spoofed packet sent to the application server, there may be a single response. However, Loop attacks are way more dangerous when the victim server who gets the response replies with its own response, which in turn is answered with another response in a loop that never ceases. For the malicious actor, it takes only a single well-crafted packet to create a Loop attack. If the attack is sent between multiple application servers, it is becoming a volumetric DDoS flood that may risk not only the application, but also the underline networks. Another interesting difference between reflected amplified UDP attacks and the Loop attack is that with Loop attack the malicious actor doesn’t control the attack lifecycle. Once the first packet is generated the Loop starts, and there’s no way for the attacker to stop it.

Reflected Amplified Attack Landscape in Azure

Since reflected amplified UDP attacks are similar to Loop attacks in their basic reflection pattern and their volumetric nature, we provide recent reflected attack landscape in Azure. As we see in the figure, UDP reflected amplification attacks account for 7% of all attacks in the first quarter of 2024.

Figure 1 – distribution of main attack vectors in Azure, January-March 2024

Out of the main attack vectors, we see reflected amplified UDP attacks, such as SSDP, DNS, NTP, CLDAP, Memcached and CharGEN as the top attack vectors.

Figure 2 – distribution of reflected amplified UDP attack vectors in Azure, January-March 2024

Azure’s Approach to DDoS Attacks

Azure employs multiple layers of DDoS protection across its network topology and OSI layers. From OSI layers perspective, DDoS attacks can manifest themselves as application and/or network layer attacks. Depending on the attacked application, different countermeasures are employed to protect against web attacks, DNS attacks, and attacks on gaming workloads. The common denominator for all protections is a DDoS network protection layer whose role is to safeguard our platform against volumetric floods, regardless of whether it’s an application attack that presents itself as a network flood, or a volumetric TCP or UDP attack.

Moreover, we apply the best mitigation at different places in the network to mitigate attacks in the most efficient manner and closest to the attack source as possible. Protections are employed at multiple rings in the network. The first ring aims to protect our peering links with other networks. When attacks risk our peering links, we mitigate them outside of our network, typically at the transit network. Multiple countermeasures are utilized to mitigate attacks outside of our network, and we select those that are most valuable for a specific scenario. The second ring is our network edge. We use inline and out-of-path DDoS pipelines to mitigate attacks that enter our network at the edge. We use various L3-L7 countermeasures to protect against network floods, web applications and DNS attacks, be it volumetric or low & slow. The next ring is our WAN network, at which we employ strategies to prevent our WAN network getting saturated by very large network floods. We employ various traffic engineering and machine learning models to detect and to deprioritize attack traffic when we need to throttle it. The last ring is in the region, where we employ an advanced mitigation pipeline who can mitigate attack vectors to protect our data centers’ infrastructure and customers’ workloads. In this ring we employ specific application and network-layer DDoS countermeasures to mitigate advanced attack vectors.

Figure 3 – Azure’s protection rings

Mitigating Loop attacks

CISPA researchers haven’t witnessed an exploited Loop attack in the wild yet. Moreover, the exact vulnerable payloads used to trigger the attacks in their research were not disclosed publicly. Yet, application vendors susceptible to Loop attacks should patch this vulnerability in their applications, and new application designs should account for this potential attack vendor in their development cycle.

As we’ve mentioned, Loop attacks may impact applications in two ways: (1) the endless loop of error messages from one application server to another may overwhelm the CPU and render it unavailable to serve legitimate client requests; (2) depending on the network topology of a Loop attack, it may become volumetric in nature hence saturate networks and network connections. Each of the protection rings described earlier will play an active role to defend against such attacks when they emerge.

We haven’t monitored Loop attacks in Azure up to date. However, we have built-in countermeasures that will mitigate Loop attacks once they become volumetric. Loop attacks are between servers, and our pipeline is able to detect spoofed sources, reflected attacks, and attacks between application servers. Specifically, for UDP-based applications, we can detect and mitigate Loop attacks for vulnerable applications, this includes DNS, Memcached, NTP, SSDP and more. We do it by monitoring communication between application servers and throttling volumetric floods. This ensures that Loop attacks, if happen, will not risk our platform, network and customers’ applications. To prevent Loop attacks from impacting application servers’ due to depleted CPU resources, application owners should patch their servers against this vulnerability.

Recommendations for Azure users

We recommend all customers with resources that are reachable from the Internet to onboard to Azure’s DDoS Protection offering. With this protection they ensure that their resources are protected against volumetric attacks, including Loop attacks. By onboarding to DDoS Protection SKUs, customers get protection tailored to their specific application footprint. This ensures we apply a DDoS policy that adheres to the exact application posture. Since each application is different, it’s crucial to apply a policy that best represents the protection for that application.

Next steps

Enable DDoS protection for your resources.
Simulate DDoS attacks to make sure your resources are protected.
Enable telemetry and logging to monitor attacks.
Utilize Microsoft Sentinel to integrate DDoS in your Security Orchestration and Response (SOAR) strategy and get alerts and recommendations from Microsoft Defender for Cloud.

Microsoft Tech Community – Latest Blogs –Read More

Parameter Identification with PINN (Physics Informed NN)

/ 2024-05-16 / Matlab

Hello,
I have a system of PDEs simulating a heat transfer system, and I have experimental data of the behaviour of the system. I am trying to use a PINN for the identification of 4 unkwnon parameters.
I am trying first to validate the model with known parameters, but I am unable to reach any reasonable parameter value. The learning rate is too slow, any recommendations?
Thanks for the help!
To run the code you will need to download the following .mat: https://cyclomed-my.sharepoint.com/:u:/g/personal/administracion_cyclomed_onmicrosoft_com/ETlg6E6520JDnBoIOv35Bq0B0o6ZCXvAvvqG3Gn1Ht3ssA?e=hKclKH
For context this is a simplified version of the PDE:

% Identify parameters of single blow equation
% 1) Fluid (u1) –> B*du1/dt + du1/dx = 1/Pe*d2u1/dx^2 + NTU*(u2 – u1) + NTU_W(u3 – u1)
% 2) Regenerator (u2) –>du2/dx = K_R*d2u2/dx^2 – NTU*(u2 – u1)
% 3) Wall (u3) –> du3/dx = R_TC*K_W*d2u3/dx^2 – R_TC*NTU_W*(u3 – u1) – NTU_W(u3 – u3_init)

% Load Experiment data
load("Test 11.mat");

%% Neural Network Definition
batchSize = 500;
dimension = 2; % X = (t,x)

% Set up neural net.
hiddenSize = 100;
net = [
featureInputLayer(dimension)
fullyConnectedLayer(hiddenSize)
tanhLayer
fullyConnectedLayer(hiddenSize)
tanhLayer
fullyConnectedLayer(hiddenSize)
tanhLayer
fullyConnectedLayer(3)];

% Create struct of parameters
parameters.net = dlnetwork(net);

% Unknown PDE parameters
parameters.Pe = dlarray(1000);
parameters.NTU = dlarray(200);
parameters.NTU_W = dlarray(1);
parameters.NTU_EXT = dlarray(50);

% Known PDE parameters
constants.B = B;
constants.K_R = K_R;
constants.K_W = K_W;
constants.R_TC = R_TC;
constants.R_Q = R_Q;
constants.T_W = u3(1,1);

%% Experiment data — Real PDE data

% Define the size of the original matrix
[rows, cols] = size(u1);

% Generate the grid for the original matrix
[X, Y] = meshgrid(1:cols, 1:rows);

% Generate the grid for the interpolated matrix
[XI, YI] = meshgrid(linspace(1, cols, batchSize), linspace(1, rows, batchSize));

% Interpolate the matrix along each dimension
UTest1 = (interp2(X, Y, u1, XI, YI));
UTest2 = (interp2(X, Y, u2, XI, YI));
UTest3 = (interp2(X, Y, u3, XI, YI));

% Select random points
[rows, columns] = size(UTest1);
X = ceil(rand(batchSize,1)*columns);
Y = ceil(rand(batchSize,1)*rows);

for k = 1 : length(X)
row = Y(k);
col = X(k);
UTest.u1(k) = UTest1(row, col);
UTest.u2(k) = UTest2(row, col);
UTest.u3(k) = UTest3(row, col);
end

% Transfor to dlarray
x = dlarray((X./batchSize)’, "CB");
t = dlarray((Y.*t(end)./batchSize)’, "CB");
X = [t;x];

% Experiment data
UTest.u1 = dlarray(UTest.u1, "CB");
UTest.u2 = dlarray(UTest.u2, "CB");
UTest.u3 = dlarray(UTest.u3, "CB");

%% Solve inverse PINN
% Training loop
avgG = [];
avgSqG = [];
maxIters = 10000000;
lossFcn = dlaccelerate(@modelLoss);

for iter = 1:maxIters
[loss,gradients] = dlfeval(lossFcn,X,parameters,constants,UTest);

[parameters,avgG,avgSqG] = adamupdate(parameters,gradients,avgG,avgSqG,iter);

if mod(iter, 1000) == 0
fprintf("Iteration : %d, Loss : %.4f n",iter, extractdata(loss));
fprintf("Iteration: %d, Predicted Pe: %.3f, Actual Pe: %.3f, Predicted NTU: %.3f, Actual NTU: %.3fn", …
iter, extractdata(parameters.Pe), Pe, extractdata(parameters.NTU), NTU);
fprintf("Iteration: %d, Predicted NTU_W: %.3f, Actual NTU_W: %.3f, Predicted NTU_EXT: %.3f, Actual NTU_EXT: %.3fn", …
iter, extractdata(parameters.NTU_W), NTU_W, extractdata(parameters.NTU_EXT), NTU_Q);
end
end

%% PINN definition
function [loss,gradients] = modelLoss(X,parameters,constants,UTest)
% X = (t,x).
% PDE
u = predict(parameters.net, X);
u1 = u(1,:);
u2 = u(2,:);
u3 = u(3,:);

% First partial derivatives
du1 = dlgradient(sum(u1,2), X, RetainData=true, EnableHigherDerivatives=true);
du2 = dlgradient(sum(u2,2), X, RetainData=true, EnableHigherDerivatives=true);
du3 = dlgradient(sum(u3,2), X, RetainData=true, EnableHigherDerivatives=true);

du1dt = du1(1,:);
du1dx = du1(2,:);
du2dt = du2(1,:);
du2dx = du2(2,:);
du3dt = du3(1,:);
du3dx = du3(2,:);

% Second partial derivatives
d2u1 = dlgradient(sum(du1dx,2),X,RetainData=true);
d2u2 = dlgradient(sum(du2dx,2),X,RetainData=true);
d2u3 = dlgradient(sum(du3dx,2),X,RetainData=true);

d2u1dx2 = d2u1(2,:);
d2u2dx2 = d2u2(2,:);
d2u3dx2 = d2u3(2,:);

% PDE residual
odeResidual = (constants.B*du1dt + du1dx – 1/parameters.Pe*d2u1dx2 – parameters.NTU*(u2 – u1) – parameters.NTU_W*(u3 – u1)).^2;
odeResidual = odeResidual + (du2dt – constants.K_R*d2u2dx2 + parameters.NTU*(u2 – u1)).^2;
odeResidual = odeResidual + (du3dt – constants.K_W*d2u3dx2 + constants.R_TC*parameters.NTU_W*(u3 – u1) + constants.R_Q*parameters.NTU_EXT*(u3 – constants.T_W)).^2;

% Compute the mean square error of the ODE residual
pdeLoss = mean(odeResidual,"all");

% Compute the L2 difference between the predicted xpred and the true x.
reconstructionLoss = l2loss(u1,UTest.u1) + l2loss(u2,UTest.u2) + l2loss(u3,UTest.u3);

loss = pdeLoss + reconstructionLoss;
[gradients.net, gradients.Pe, gradients.NTU, gradients.NTU_W, gradients.NTU_EXT] = dlgradient(pdeLoss,parameters.net.Learnables,parameters.Pe, parameters.NTU, parameters.NTU_W, parameters.NTU_EXT);
endHello,
I have a system of PDEs simulating a heat transfer system, and I have experimental data of the behaviour of the system. I am trying to use a PINN for the identification of 4 unkwnon parameters.
I am trying first to validate the model with known parameters, but I am unable to reach any reasonable parameter value. The learning rate is too slow, any recommendations?
Thanks for the help!
To run the code you will need to download the following .mat: https://cyclomed-my.sharepoint.com/:u:/g/personal/administracion_cyclomed_onmicrosoft_com/ETlg6E6520JDnBoIOv35Bq0B0o6ZCXvAvvqG3Gn1Ht3ssA?e=hKclKH
For context this is a simplified version of the PDE: