Hi 

The short a definite answer: yes, neural networks can take weather variables (temperature, wind chill, humidity) into consideration, and also their lags (not everything is contemporaneous). Indeed neural networks are particularly well suited to forecast load demand as the relationship between weather variables and load is non-linear and the modelling is much more adequate than with linear regression.

BTW: the question of "normalising the input" is confusing´the terminology a bit: in electriciy demand forecasting you often refer to normalise the demand by deseasoanalising (i.e. removing weather effects); in Neural Networks nromalisation refers to scaling the input variables into a suitable range that the algorithm can learn from (i.e. [0; 1] or [-1, 1]).

So yes, normalise the inputs in a neural network training  sense, but do not preprocess / deseasonaliose the data as this woudl get rid of the nonlinear interactions between weather and load. 

Hope this helps, Sven

you can use ANN for long/short-term load forecasting only based on weather data, but  agriculture and residential sections demands are more depends on the weather conditions not industries.

Hi, 

I worked on these issues in the past, I have used NNs to forecast building electricity demand using several predictors: temperature, calendar data (time of the day, day of the week), people into the building, etc. 

First, you need to answer the following question: is the demand affected by weather? The answer is not (always) trivial: e.g. during Winter if the buildings use gas heating (like is common in Italy) you can't see any direct relationship between demand and temperature. 

Once you've realized that weather variables can be useful you can use them as predictors paying attention to the normalization of variables and avoiding correlation among them. 

Fuzzy contro;er would be better

Hi 

The short a definite answer: yes, neural networks can take weather variables (temperature, wind chill, humidity) into consideration, and also their lags (not everything is contemporaneous). Indeed neural networks are particularly well suited to forecast load demand as the relationship between weather variables and load is non-linear and the modelling is much more adequate than with linear regression.

BTW: the question of "normalising the input" is confusing´the terminology a bit: in electriciy demand forecasting you often refer to normalise the demand by deseasoanalising (i.e. removing weather effects); in Neural Networks nromalisation refers to scaling the input variables into a suitable range that the algorithm can learn from (i.e. [0; 1] or [-1, 1]).

So yes, normalise the inputs in a neural network training  sense, but do not preprocess / deseasonaliose the data as this woudl get rid of the nonlinear interactions between weather and load. 

Hope this helps, Sven

Hello,

as Mr. De Felice stated, it is quite important to identify the weather dependent loads in your grid. For electrical heating our experience in a real world scenario is that the total needed power on a daily bases correspons almost linear with the temperature. Thus, for our demand-response project, sophisticated approaches like ANN were not necessary at all.

Best regards

Sorry, but I don't understand the question: "separate input/output". In our departmenet we applied ANN for load forecxasting in real utility environment (more than twenty years ago) taking into account temperature, humidity, darkness of a day and/or any additional weather parameters (acord to utility' need) with good results. Our programs were used  for few years up to the moment of reorganizing utility.

Hello,

I certainly agree with others that mentioned that you can use NN. In the other hand, based on my experience (price forecasting) Support Vector Regression works better than NN. Usually NN works well in the training data but sometime they fail in testing data. Also, they converge to local optimal points. Still NN is a good method to try (not that SVR is perfect... there are also other issues with it).

Regards,

Felipe 

Thank you for all the responses, certainly very very helpful.

Gerro

There are many good answers here,

may I also suggest that you investigate recurrent neural networks in this research.

The approach to use a ANN would definitely work however we have been looking into Bayesian networks for more granular tuning. A weather forecast seemingly fits well within a a probabilistic model.

NN works very well for such kind of problems. One important issue is that enough set of data with varied patterns should be available for training purpose.

Yes! I have used for electricity as well as heat demand forecast. And it worked as expected.

ANN are very well suited for your kind of applications. In real world applications you have to carefully consider what kind of application you want to build. Modelling the current behavior of your power supply is one thing. Modelling predictions is a different thing. It might be helpfull for you to look into time delay or recurrent ANNs for your kind of application. Another point is to carefully select the system variables that you observe for modelling, because your main interest is on predicting the energy demand based on weather forecast and not to build a model of the closed loop control circuit.

This paper explains the usage of SWM . "Electricity Load Forecasting for Residential Customers: Exploiting Aggregation and Correlation between Households"  good reference to start 

http://infoscience.epfl.ch/record/188682