This has to be studied by defining in your particular context a network of nodes.

Each node has three functionalities:

-network (link, transmit input or output, eg NN)

-data (retrieve, store, etc)

-compute (DL, ML, etc).

Once you have done this scheme, or description, everything can be clearly discussed and optimised

Hi Madhusmita Mishra

You can try Matlab artificial neural network tool, it is quite useful for forecasting.

However, you need a lot of historical data.

Madhusmita Mishra ,

I would first establish what is the main goal of your study

Most publications that I have seen in ML for net load forecasting and line loss prediction are from computer scientists applying algorithms to test their performance against a know data set.

Things in reality are a lot different in real life scenarios

Lets take the problem of net load forecasting

net load forecasting can be tricky since getting accurate data requires substantial time and there can be variations from grid to grid such as:

• Average load vs peak load between regions of the grid
• Topology of the grid (e.g. sole provider vs shared generation)
• where are the measurement units (E.g. generation vs distribution, etc.)

Such analysis should be done before even doing any kind of ML. Once the study has been done several algorithms should be tried to see how they perform against the actual data. the particulars of the data will dictate the algorithm ( what is applied to one data set does not guarantee effectiveness under all circumstances

I would also address your statement : "intelligent robots be implemented "

Robots can be used mainly as agents for data gathering, monitoring and performance of the system under study. the robot can use AI (AI is not necessary, since other more rudimentary programming can be used) to find:

• Particular paths on the system
• Evaluate whether there is damage on the line, etc.

The takeaway is:

Do not be biased in choosing a particular approach from the start, the topology and circumstances of the system of the study andthe data will dictate the best solution (e.g. for algorithms, if the data collected is binary you will need a neural network with binary inputs. If the data is non linear then that will eliminate other algorithms, etc.)

Hope this helps