Abstract:Electricity consumption has increased exponentially during the past few decades. This increase is heavily burdening the electricity distributors. Therefore, predicting the future demand for electricity consumption will provide an upper hand to the electricity distributor. Predicting electricity consumption requires many parameters. The paper presents two approaches with one using a Recurrent Neural Network (RNN) and another one using a Long Short Term Memory (LSTM) network, which only considers the previous electricity consumption to predict the future electricity consumption. These models were tested on the publicly available London smart meter dataset. To assess the applicability of the RNN and the LSTM network to predict electricity consumption, they were tested to predict for an individual house and a block of houses for a given time period. The predictions were done for daily, trimester and 13 months, which covers short term, mid-term and long term prediction. Both the RNN and the LSTM network have achieved an average Root Mean Square error of 0.1.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: **How to accurately predict future electricity consumption** in order to help electricity distributors better cope with the increasing electricity demand. With the exponential growth of electricity consumption in the past few decades, electricity distributors are facing a huge burden. Therefore, the need to accurately predict future electricity consumption has become crucial.
### Specific problems:
1. **Short - term, medium - term and long - term electricity consumption prediction**: The paper aims to develop models that can perform short - term (such as one day), medium - term (such as three months) and long - term (such as 13 months) electricity consumption prediction.
2. **Balance between complexity and accuracy**: Electricity consumption is affected by multiple factors (such as weather, number of residents, equipment type, etc.), making the prediction model complex and difficult to stabilize. Therefore, the paper attempts to use time - series methods, especially deep - learning models, to simplify feature selection and improve prediction accuracy.
3. **Prediction of individual houses and housing blocks**: It is necessary not only to predict the electricity consumption of a single house, but also to predict the overall electricity consumption of a group of houses (i.e., a block).
### Solutions:
The paper proposes two time - series prediction models based on deep learning:
- **Recurrent Neural Network (RNN)**
- **Long - Short - Term Memory Network (LSTM)**
These two models only use past electricity consumption data to predict future electricity consumption and have been tested on the publicly available London smart meter data set. The experimental results show that RNN and LSTM perform well in short - term, medium - term and long - term predictions, especially compared with the traditional ARIMA, ANN and DNN models, RNN and LSTM perform more excellently in medium - term and long - term predictions.
### Key formulas:
- The recursive formula of RNN is:
\[
y_t=\tanh(W_r y_{t - 1}+W_n x_t)
\]
where \(y_t\) is the current output, \(y_{t - 1}\) is the output at the previous moment, \(x_t\) is the current input, and \(W_r\) and \(W_n\) are weight matrices respectively.
- The structure of LSTM is more complex and contains three gating mechanisms (input gate, forget gate and output gate) to control the flow of information. The specific formulas are as follows:
- Input gate:
\[
i_t = \sigma(W_i[h_{t - 1},x_t]+b_i)
\]
- Forget gate:
\[
f_t=\sigma(W_f[h_{t - 1},x_t]+b_f)
\]
- Output gate:
\[
o_t=\sigma(W_o[h_{t - 1},x_t]+b_o)
\]
- Cell state update:
\[
\tilde{C}_t=\tanh(W_C[h_{t - 1},x_t]+b_C)
\]
- Final cell state:
\[
C_t=f_t\odot C_{t - 1}+i_t\odot\tilde{C}_t
\]
- Final output:
\[
h_t=o_t\odot\tanh(C_t)
\]
Through these models, the paper shows the potential of deep - learning methods in electricity consumption prediction, especially the advantages in dealing with long - time - series data.