According to my understanding online means the analysis gets done when the data is captured, i.e., Real time analysis. This approach results in speedy result generation.

In case of offline, the data is captured first and later on the analysis takes place, which may be a time consuming task.

According to my understanding online means the analysis gets done when the data is captured, i.e., Real time analysis. This approach results in speedy result generation.

In case of offline, the data is captured first and later on the analysis takes place, which may be a time consuming task.

Both cases are sometimes a necessity, e.g. interferometry in astronomy. Interferometry allows to get a high resolution image by superposition of data of more (at least two) distant telescopes. In optical interferometry, due to very short wavelength (very high frequency and fine phase issues), one have to do it in real time, two optical telescopes must be connected by a very precise optical path and the optical images interfere the same time. In radioastronomy you can have separate off-line radio-images records and "interfere them" off-line by the computers later, however you have to record them with very precise atomic clocks that allow necessary synchronization. Thus you can provide radiointerferometry from radiotelescopes located on distant places on Earth surface, even on orbit (e.g. Spektr-R / Radioastron).

In system identification, when the analysis is online, the parameters of the identified system can be updated after and during each experiment. But in offline analysis, the system is identified based on all gathered data from experiments after finishing all experiments and the identification process happens only once (while it could contain lots of loops).

In offline identification, the identified system is a time-invariable system with constant parameters. But in online method the result is usually a structure for the system which contains time-variable parameters which are updated and got more optimized after or during each experiment. This feature allows the system to have different and nonlinear behaviors in different conditions which aren't anticipated based on system inputs. Online method is usually used when you want to control a time-variant system which is not identified, but you need to assume a model to control it, so the controller should update the assumed model by each new data, because the behavior can't be foreseen by a constant model.

Kianoosh Karimi ..Thank you Kianossh....I am using time series data to forecast future electricity demand using A.I algorithms like machine learning.I have historical electricity data but my model will predict the future demands of electricity using the historical inputs.I have set optimization algorithm to minimize the prediction error (it repeats itself until goal error is achieved).So,the whole scenario can be considered offline or Online?

Muhammad Ali ... you're welcome

If your historical inputs and output(s) are a particular set of data which your model is completely identified based on them and it is not needed to adjust the model to every new data added to your data base, so the proposed method is offline.

There is another point about prediction error; if the model will predict the future demands only using historical inputs and the variation range of your historical inputs covers all probable future input values, the procedure in indeed a simulation procedure and optimization algorithm should be set to minimize the simulation error. Prediction error minimization is used when there are some probable input values which the system hasn't experienced them in your historical data and you want to define a model which can predict the output when those new input values are exerted.

Online processing The term 'online' processing is the term used to describe when a user is connected to a computer or network (they are logged in) and are processing the data files at the same time as using the input, output and storage devices.

Offline processing To help speed up data input and output into computer systems, offline processing was 'invented'. This removes the need for relatively slow input devices. Instead, data is stored in files on a high-speed data storage device. This could be done in a number of ways. For example, an operator might use a different computer system to simply type in data from paper forms into a file, perhaps using a key-to-disk system or if the data was in a machine-readable form (such as OMR sheets), it could be read automatically and stored in a file. Then, when the computer is ready, it can get all the data it needs from these files on the high-speed storage device rather than from relatively slow input devices.

This is described as offline processing because the main computer doing the processing is not immediately controlling and reading the data from its input devices. The data is being prepared and stored away from that computer on a high-speed storage device and is then made available as necessary.