How To Get ANN Training Data From SIMULINK ? (Part-4) | Dr. J. A. Laghari

[@janardhangurram3413]

Sir, You are truly I have no words to praise U , This series of ANN lectures made my understandings next level .Thank you vey much.

[@DrJALAGHARI]

Thank you so much for the compliments.

[@CarlosHenriqueDomiciano]

I enjoy this class, congratulations. I complete ANN's series. Thank you very much.

[@DrJALAGHARI]

Congratulations for watching the whole series.

[@sds2908]

very great lecture! I have learned something new from your ANN lectures. keep it up, Sir.

[@DrJALAGHARI]

Thank you so much for the compliments. Do share it within your circle too so that they may also get benefit from this video and channel.

[@mahboobulhassan5560]

Thank you very much Sir for creating such awesome videos. I learned a lot by watching series of lectures on Load Frequency Control. I am also working on the integration of Renewable energy sources and looking forward you to making such kinds of videos.

[@DrJALAGHARI]

Thank you so much for the compliments. Yes sure.

[@j.ymaipan-uku9713]

Thanks a lot Prof. you are indeed a saviour. Jazakallahu Khairan

[@DrJALAGHARI]

Thanks for the compliments.

[@davidezekiel685]

I have gotten the corrections. I will do as instructed. God bless you great Prof. It God bless you abundantly Sir and remain blessed

[@DrJALAGHARI]

Thank you so much for the compliments and wishes. Stay blessed

[@QusayALHAMD]

Very Great work Dr , inspiring , however , would you please shed some lights if you had implement the ANN for the power transformer fault detection before or not ? I have searched the channel but I couldn't find

[@DrJALAGHARI]

Thanks for the compliments. No. I have not used it. But, the idea is same.

[@muhannaddrakalsebai4839]

Thank you so much doctor, It was really useful. JAZAKA ALLAH.

[@DrJALAGHARI]

Thanks for the compliments.

[@robbyiqbalulmaulidiTE]

Thank you very much, may Allah always bless you sir

[@DrJALAGHARI]

Thanks for the compliments.

[@franklinobakpolor4584]

Thank you very much for the series videos, it has been of great help.

[@DrJALAGHARI]

Thanks for the compliments.

[@franklinobakpolor4584]

@@DrJALAGHARI Please sir I sent you a message via email, I will appreciate it if you take a look at it. Thank you sir

[@DrJALAGHARI]

@@franklinobakpolor4584 Thanks for your message. I have sent the reply. Kindly check it.

[@farahqatrunnada7747]

Thank you so much for this helpful video. May I ask, what if I want the ANN to generate the PID parameters. What Input and Target data is needed to train the ANN? I would be very helped with your answer. Thank you in advanced.

[@DrJALAGHARI]

Thanks for the compliments. If you have simulink model of PID. Then follow the above strategy and use three output parameters of PId controller as output by using multiplexer.

[@vitordesanunes1206]

Really much appreciated, this helped me a lot. Greetings from Brazil.

[@DrJALAGHARI]

Thanks for the compliments.

[@gianlucaformiglio980]

Good evening Prof., how can I use the neural network to tune the gains of the PID controller?

[@DrJALAGHARI]

Thanks for the comment. You may simulate the PID controller based model and get the training data then use the data for training and testing of ANN.

[@Electrical.Engineer730]

Thank you so much for the video my dear respected sir

[@DrJALAGHARI]

My pleasure.

[@mwaseem2625]

Thanks Sir, I learned a lot by watching series of lectures on ANN training. But sir I am facing a problem in training of ANN for PSS and FACTS controllers that are used for Transient Stability improvement of the Power Network.

[@DrJALAGHARI]

Thanks for the compliments. May I know what problem you are facing?

[@mwaseem2625]

Sir I am training ANN for PSS, for that taking data to workspace using 'to workspace'tool, but during training ANN that data is not applicable or error occurrs

[@DrJALAGHARI]

@@mwaseem2625 Which Matlab version you are using?

[@mwaseem2625]

It's 14a

[@rihanabanu9985]

Thank u Dr. Laghari for ur valuable lectures...It would be very helpful if u do let us know and teach us how to tune PID controller parameters using ANN for first order or second order process....

[@DrJALAGHARI]

Thanks for the compliments. The procedure of ANN training remains same for all type of process. Only input and output data is required which is the most important part of the ANN training.

[@rihanabanu9985]

@@DrJALAGHARI Thank u for ur reply....How to create a neural model with four inputs like e, de/dt, r and y and 3 outputs Kp, Ki and Kd....In the video shown there is only one input e(error) and one output u(controller output)....Pls can u suggest....

[@DrJALAGHARI]

@@rihanabanu9985 Thanks for the message. De/dt can be obtained by using the derivative function block in the simulink model. By this way, one will be the original input, and other will be its derivative input. Then you can take the multilpexer block to combine these two inputs and send it to workspace. I couldn't understand r and y ?

[@rihanabanu9985]

@@DrJALAGHARI Assalamu alaikum sir.....Can u pls post ur mail id

[@DrJALAGHARI]

@@rihanabanu9985 [e-mail] is my email address.

[@farhannajihah4474]

Very good lecture series Sir! Thank you.

[@DrJALAGHARI]

Thanks for the compliments.

[@rameshdeepa486]

Thank you very much sir, how to find setting time, peak over/under shot

[@DrJALAGHARI]

Thanks for the comment. You have to take all data from the simulink. The data itself contains the undershoot, overshoot values.

[@SUHAILAhmed-zr3zq]

Thankyou Sir for helping student's❤

[@DrJALAGHARI]

Thanks for the compliments.

[@bhoopendrasinghrathore1460]

Respected Sir Could you please help me in my topic of ANN MPPT based Wind turbine? How to implement ANN with MPPT for fixed speed generator wind turbine?

[@DrJALAGHARI]

I have not worked on it. But steps are same. You may follow these steps and implement it.

[@hassaniskandarani1345]

Thank you doctor for the video, but what should I do if I want to connect NN+PI to ameliorate the PI output instead of replacing the PI, I'll appreciate any help you can give me.

[@DrJALAGHARI]

Thanks for the comment. May i know what is the advantage of using both ann and pi in one model?

[@hassaniskandarani1345]

@@DrJALAGHARI yes Sir, I am working on a NN that is used in parallel with a non robust PI controller, the goal is to get a better response when using the NN+PI than the PI alone, just like this paper (Artificial Neural Network Based Adaptive Control of Single Phase Dual Active Bridge With Finite Time Disturbance Compensation) in the figure 5, but I don't know how to implement the adaptive control for the neural network in MATLAB, I saw an adapt tab in the nntool I don't know if it helps, if you have any idea about it I'll be thankful to you. I replied to you like an hour ago but the comment got deleted because it contained an IEEE link.

[@DrJALAGHARI]

@@hassaniskandarani1345 ii am not sure about combining both. You may search for many research papers to fully understand this idea.

[@hassaniskandarani1345]

@@DrJALAGHARI okay doctor thank you 👍🏻

[@niharikasawhney2237]

thankyou for your video sir, for fault detection using ANN m not able to think what should be taken as input, output and target. please help sir, thankyou in advance

[@DrJALAGHARI]

I have not done fault detection. but in my opinion, you have to put the input data for before fault and output data after fault in order distinguish or detect the fault location.

[@niharikasawhney2237]

@@DrJALAGHARI sir please make a video on this topic as ANN is new research for me. Please sir

[@DrJALAGHARI]

@@niharikasawhney2237 ok. I will try to make video on this topic as soon as possible.

[@niharikasawhney2237]

@@DrJALAGHARI thanks alot sir. I am doing my thesis work on this. I would be really grateful for your help

[@aungonrafin6422]

Assalam Walaikum sir, i tried to replace the PI controller with ANN in a buck boost converter circuit. The training was successful, but the problem is ANN is behaving exactly like the PI controller. Same amount of overshoot and settling time. What can I do to get a better and smoother response with ANN controller? I also tried to tune the PI controller first. but the tuner couldn't linearize the model. Please help me.

[@DrJALAGHARI]

Thanks for the comment. You can reduce the overshoot values by changing them with lower values and then train. Now it will give smaller overshoot

[@sravankumar1263]

Thankyou very much for the video and useful information sir. Sir please tell me how to get the input data for ann inorder to perform load flow analysis sir..

[@DrJALAGHARI]

Thanks for the compliments. for this purpose, try to follow any research paper closely related to this topic for better understanding.

[@ShalakaSitre]

Can you suggest how ANN can be used for harmonic mitigation in multidrive system? based on filter of PID controller in Active front end?

[@DrJALAGHARI]

The procedure of ANN implemntation remains same for all systems. In your case first simulate the model with PID controller based filter, get the input and output data of PID controller. Shift it to workspace, then train it, and get back again in the simulink model.

[@alexpilloni]

Nice talk! :-) Just a curiosity: what is for you the advantage of using ANN for implementing a controller? I mean... in general and even for more complex scenarios than the one of this example. My point is: it seems every ANN-based controller needs, to be trained, a well-tuned traditional controller. Thus, why one should prefer ANN-based controller? It seems we are doing the job twice... and the final controller works further worse the PID, due to the lack of the integral action in the ANN-controller... Moreover note that the complexity burden of the ANN-based controller is much higher than the PID😀

[@DrJALAGHARI]

Thanks for your valuable comment. PID controller has the limitation of the parameter tuning and also it does not provide an optimum response for every new changing condition. furthermore, for every new situation, it tries to give the response at that time due to which its computational time is quite large. This results in a higher overshoot, and a large settling time. On the other hand, ANN requires training once. After the successful training, It will provide the best response for every new condition. Furthermore, as its response is already given, therefore, it will not take time to calculate the best response for every new condition. due to this, its computational time is smaller. This results in smaller overshoot, and lesser settling time.

[@alexpilloni]

@@DrJALAGHARI I mean, take a look to 8:30 the ANN-controller has 100% of overshoot, very bad time response and worse steady state error... Moreover, it is unclear why you say the ANN-tuning requires training once. Since your ANN tuning looks based on a given PID response, the ANN tuning cannot perform better then your original PID. Possibly... are you referring (maybe) to a different tuning method than the one you show in this video? Do you have any reference? ps. Regarding the complexity... a standard PID accounts only 3 registers, whereas, how many neurons do you need for implementing a ANN-based controller?

[@DrJALAGHARI]

@@alexpilloni Thanks for your valuable feedback. I understood now your point of view. The main aim of the video is how to get ANN training data from Simulink and use it for training or testing. I have not properly trained and tested the ANN. I have just shown an example of how to obtain the data.

[@abdulazizalbarazi1346]

Assalamualaikum sir, I have a problem with the Input and Output data size. At 4:05, it is shown in the workspace that the size is 300006×1, but for me, the size of the data is only getting to 916×1, which is a VERY LARGE difference. I have done exactly what was shown in the video, the response curve and everything is exactly the same. I could not find a solution for that.

[@DrJALAGHARI]

Thanks for the compliments. it depends upon the step size settings in the matlab Simulink. you can modify it from the simulation configurations.

[@atiyahwaael1955]

thank you doctor for this lectures, my question how I can discriminations between types of harmonic in sinewave or power generator by using ANN

[@DrJALAGHARI]

For this purpose, first you have to use the signal processing techniques such as FFT, DFT, wavelet etc then take that data in order to distinguish using ANN. Better to follow any research paper.

[@waaelabdulhassan4982]

Thank you Doctor for reply , I already using FFT for this purpose but my question where I can put block of ANN 1-Instead of differential relay Between input and output current 2-before FFT block 3-After FFT block Or there are another option Dear Dr possible to explain that

[@DrJALAGHARI]

@@waaelabdulhassan4982 use the ANN after the FFT block .

[@waaelabdulhassan4982]

Thank you for clarification , I will do and update you about situation

[@naveenkumar-zy4po]

thank u sir I tried in Matlab sir. I wrote the Matlab program for ANN What you have shown in the video. but I am getting errors, sir. please solve my problem sir.

[@DrJALAGHARI]

Thanks for the comment. Download the matlab program from the link given in the description section of the video.

[@debdeepsaha5120]

Very much informative video. Great Job Sir !! I would like to take a help from you on ANN !!

[@DrJALAGHARI]

Thanks for the compliments.

[@mashitahcherazali7895]

Thank you for great explanation Sir. Sorry Sir, I really need your help. I try to replace PID controller(2 PID controller, decentralised type) with ANN controller (2 ANN controller, decentralised type). I used m-file and gensim command like you explained. However, I used 2 m-file since I need two ANN controller. I'm able to get 2 ANN controller with R=0.99984 and R=0.98222 for each. When I used the obtained ANN model block at simulink and try to run the model, it give an error. The error stated "at time 0.0 using the TrustRegion-based algorithm due to one of the following reasons: the model is ill-defined i.e., the system equations do not have a solution; or the nonlinear equation solver failed to converge due to numerical issues.".Meanwhile, the same model can give good result with PID controller. It is due to R not equal to 1 or why? Can you help me sir? Thanks in advanced

[@DrJALAGHARI]

Thanks for the comment. I think it should work. R above 0.9 is acceptable. this is not due to R. but may be due to another reason.

[@alrawdasuliman6649]

Sir, am getting this error ( Output data size does not match net.outputs{3}.size.), using one PID controller, my second question , my model has more than PID controller , how to deal with , in case using ANN??

[@DrJALAGHARI]

For multiple pid controllers you have to use multi ANN controller. Therefore at first you have to replace one pid controller with ANN controller then second one.

[@alrawdasuliman6649]

@@DrJALAGHARI Jazak Allah Sir.

[@alrawdasuliman6649]

@@DrJALAGHARI sir, still getting the error "Output data size does not match net.outputs{3}.size," am using input data matrix (13198*1), if i transpose the matrix is it going to effect my signal to controller ? or any other solution pls, and yes i have changed the ANN input and output to [1,8,1]

[@niharikasawhney2237]

sir please make a video on fault detection and location using ANN

[@DrJALAGHARI]

Thanks for the comment. Ok i will try to make it as soon as possible. But it is not easy. I have to apply first this technique. Let see how much time it takes.

[@niharikasawhney2237]

@@DrJALAGHARI thankyou sir

[@kamgangblaise2645]

Thanks so much for the video. I have a question. What if the input to the ANN is more than one, how do you model it in Simulink? The model you presented has just one input in your Simulink. I am asking because my Simulink model has 3 inputs so I am having some difficulties.

[@DrJALAGHARI]

Thanks for the comment. You can adopt same procedure for more than one output. In that case, output columns will be two or more depending upon the number of outputs.

[@kamgangblaise2645]

@@DrJALAGHARI thanks. My output is one. instead i have 3 inputs

[@DrJALAGHARI]

@@kamgangblaise2645 The procedure will be same. Input data will have three columns.

[@davidezekiel685]

Thank you very much Sir for such informative tutorial, simple, clear, timely and informative! Sir, I have a question and kindly help me!! I am working on simulating a MIMO (Actually 2-input 2-output, i.e. TITO) system. I tried using the code from this video and the example based on Simulink design, but i could not simulate (i.e. get the system to run). Sir, how can I use an ANN controller for this TITO system simulation, or what can i do additionally to make the system run? Cheers

[@DrJALAGHARI]

Thanks for the comment. The procedure is same. I don't know what problem you are facing?

[@davidezekiel685]

Thank you great Prof. You will live long on this earth to accomplish all that God wants u to accomplish for Him and for humanity. Sir, I sent you a message to your inbox. Can you kindly check and help me. I am at a crossroads with no solution to my simulation. I believe you are a God-sent to me. Thanks

[@davidezekiel685]

Thanks in advance

[@DrJALAGHARI]

@@davidezekiel685 Thanks for the compliments. ok I will check and let you know if I can do anything

[@davidezekiel685]

@@DrJALAGHARI Thank you Sir. Very much appreciate Sir

[@abhishekraj2336]

Hello sir, I want to divide ratios for training, validation, and testing. what could be the codes for that. well, I am using this net.divideParam.trainRatio = 0.7; % training set [%] net.divideParam.valRatio = 0.15; % validation set [%] net.divideParam.testRatio = 0.15; % test set [%] But this doesn't work. It would be great if you would help in this.

[@DrJALAGHARI]

it should work. in other case, just use nntool gui, and perform the same by dividing the data into separate files and upload them separately.

[@nadiabensiali8500]

Good moorning sir I saw your video on ANN training it was very helpful . i would like to thank you for all your videos .i understand how to do training if we use boost or buck converters. I done simulation files of pv system with P&O for MPPT. I would like if you could tell me how can i do training of irradiance and temperature if i used them as inputs and duty cycle as out put . we must have mathematical equations between T,G and Vpv,Ipv(corresponding to Pmax). some works took Ipv and Vpv as inputs and Vref as out put . I will be very greatful if you could help me to write matlab instruction to do training for MPPT in PV system. i need only how to do training. thank you very much

[@DrJALAGHARI]

Thanks for the compliments. It depends upon you which parameters you are going to choose for input and output. Try to find any paper closely related to this topic. It may help you. Or you may consider any data which provides the best results. In order to get the data, run the simulation and obtained the inpit and output training data by following the procedure as mentioned in the video.

[@Ermohit25]

sir can you help me regarding direct torque control of pmsm using artificial neural network?

[@DrJALAGHARI]

Thanks for your message. What help you need?

[@shraddhagajbhiye4513]

Thank you very much sir

[@DrJALAGHARI]

Thanks for the compliments.

[@vinayshetty3059]

Can we make use of ann for placement of dg and selection of candidate node sir for an ieee distribution system and what must be my inputs and outputs to ann? Thanks in advance sir

[@DrJALAGHARI]

Thanks for the comment. In my opinion, for placement of DG and sizing of DG, optimization techniques are used.

[@vinayshetty3059]

@@DrJALAGHARI sir cant we make use of ann

[@DrJALAGHARI]

@@vinayshetty3059 in my opinion, better to search for research paper. It may be possible.

[@shaimbiswas8840]

Assalamu alaikum Sir. I have some questions. As in this video, the regression is .0966 which is less than unity. But the criteria for successful ANN training is that regression must be equal. And the training data is not on the straight line. What measures should we take for successful ANN training to improve regression to unity and fit the training data on straight line?

[@DrJALAGHARI]

Thanks for the comments. I have discussed all the possible steps to obtain the regression equal or close to unity in the third part of this ANN series. Kindly have a look on it.

[@shaimbiswas8840]

@@DrJALAGHARI Thank you Sir for your kind reply. I have watched that part. Then , I have to arbitrarily change these five parameters to get best suited straight line? Sir , I have a confusion, do we need to process raw data before feeding in ANN (such as normalization,removal of outliers)? if yes, then when and how? sir, may be it's a very silly question. actually I am a new learner. Sir, it would be kind enough if you give me your email. I want to get connected with you.

[@DrJALAGHARI]

@@shaimbiswas8840 Thanks for the comment. Processing raw data or normalization is an optional thing. Suppose, you tried all efforts and steps but still not succeed to get regression equal or close to unity. Then processing or normalization may be helpful sometime in getting regression equal to 1. My email is [e-mail]

[@shaimbiswas8840]

@@DrJALAGHARI I can't thank you enough SIr. This help means a lot to me. Jazakallah khairan

[@DrJALAGHARI]

@@shaimbiswas8840 my pleasure. Stay blessed.

[@niharikasawhney2237]

sir can you please make a video on fault detection and classification using ANN

[@DrJALAGHARI]

The procedure for fault detection using ANN is same as I have discussed in this series.

[@niharikasawhney2237]

@@DrJALAGHARI sir can you guide through the process. i will be very thankful

[@DrJALAGHARI]

@@niharikasawhney2237 in simple, simulate the fault in the simulink. Transfer the whole voltage or current data to the workspace and train that data using ANN. Then bring back the trained ANN in the simulink. Run the simulation and compare the answer.

[@niharikasawhney2237]

@@DrJALAGHARI thankyou so much sir. I will try and let you know. Thankyou

[@greddyswetala8752]

Thank you very much on such informative tutorial, simple and clear! Sir, I have a question and kindly help me!! If the ANN controller is learning from PID controller (by fitting) is there any possibility of improving the response such that the ANN performs better than PID?? Thank you Sir

[@DrJALAGHARI]

Thanks for the compliments. Yes. The response of ANN will be better than pid due to the reason that PID controller always try to modify the response at every instant of time. Due to which it takes time. On the other hand, ANN once trained, can provide the fastest response due to which its response will be faster.

[@greddyswetala8752]

@@DrJALAGHARI Thank you for clarification Sir

[@tahirusman1129]

Thank you. Masha Allah

[@DrJALAGHARI]

Thanks for the compliments.

[@shafqatabbas7967]

What we do if the size of the input and output data not in equal

[@DrJALAGHARI]

Thanks for the comment and query. I could not understand your query. can you please elaborate it further?

[@okunnubiolalekan7271]

Hello Dr thanks for the lectures. Please how can I collect data from different types of faults( e.g L-G, DL-G fault, three phase fault) to workplace for training.

[@okunnubiolalekan7271]

How will I get the input and output from the fault block

[@DrJALAGHARI]

@@okunnubiolalekan7271 Thanks for the compliments. For faults, you have to measure all theee phase currents and neutral currents . I will upload a video on this topoc may be this Monday or next Monday for your help.

[@okunnubiolalekan7271]

@@DrJALAGHARI thanks sir, please send the link to download the MATLAB program here for ANN

[@DrJALAGHARI]

@@okunnubiolalekan7271 in the first part of ANN, the link is given in the description. You may download it from there

[@okunnubiolalekan7271]

@@DrJALAGHARI OK sir thanks

[@mahmoodhamid-9647]

thank you so much

[@DrJALAGHARI]

Thanks for the compliments.

[@Dieu_Seul_Suffit__PJG]

good

[@DrJALAGHARI]

THanks

[@FaraedoonWaly]

Asalamualaikm Dear Dr J. A. Laghari. Thank you very much for your series lectures based on ANN. Please could you make series lectures for Model Predictive controller (MPC). I mean the programing and basic structure of MPC. there has a MPC block on matlab simulation but it is not efficient. If you are able to help as to understand MPC in matlab code that would be great. Thank you god blessing you inshallah

[@DrJALAGHARI]

Wa'alaikum salaam thanks for the compliments. Actually personally I have not worked on MPC. Therefore, I am not expert in MPC.

[@-workshop6370]

من فضلك انا احتاج مساعدة في Ann

[@DrJALAGHARI]

JazakALLAH

[@-BommakantiRanjith]

Thank you so much sir

[@DrJALAGHARI]

Most welcome