A smart grid is a modern electrical grid with numerous enhancements such as ability to sense and gather information through the use of communication technologies resulting in improvement in the efficiency, reliability and security of the power grid. Smart metering has brought upon immense improvement to the current billing and demand estimation models central to which is the introduction of real-time metering. A resulting benefit of this is the hugely improved power demand forecasting based on such meter measurement, which affects the power generation scheduling and power storing for a future period through prediction of the power demand in that period using past usage and load demand data.
Power demand forecasting is important for both power companies and power consumers. In general, the forecasting results have different interpretations when applied to the aggregation and the individual. The aggregation forecasting, which is to predict the power demand of a number of consumption units, e.g., the apartments within an area, is more meaningful to power utility companies. Based on the aggregation demand forecasting results, they can allocate proper resources to balance the supply and demand or adjust the demand response strategy such as dynamic pricing to shape the load so as to avoid the infrastructure capacity strain. On the other hand, individual power demand forecasting assists in the anomaly detection task in the smart metering system. Anomaly detection detects the abnormal meter measurements caused either by the unexpected meter failure or the deliberate meter manipulation by identifying those measurements that do not present a conformation to the predicted/expected values. Moreover, under dynamic pricing strategy, individual power forecasting also provides power consumers with their expected power consumption and cost in a future period, so that they can optimize their usage schedule accordingly to achieve a lower cost.
Power demand forecasting has been widely studied due to its significance in power industry. The existing works can be generally classified into two categories, i.e., classic statistical models and modern machine learning algorithms.
Time-series modeling is a very popular statistical model that is used to capture the time-series characteristics of power demand, e.g., ARMA [14,19], ARIMA [2,7]. Hong et al. [16] adopt multiple linear regression to model hourly energy demand using seasonality (regarding year, week, and day) and temperature information. Their results indicate that complex featuring of the same information results in a more accurate forecasting. Fan and Hyndman [8] use semi-parametric additive model to explore the non-linear relationship between energy usage data and variables, i.e., calendar variables, consumption observations, and temperatures, in the short-term time period. Their model demonstrates sensitivity towards the temperature. Recently, conditional kernel density estimation is applied to power demand forecasting area [4] which performs well on dataset with strong seasonality. Time-series models are based on the assumption that the future power demand has the same or similar trend and distribution as the observed history. However, the power demand in reality is influenced by many factors in various ways. Therefore, it is essential to take these influential factors into consideration.
There are three major machine-learning algorithms used in demand forecasting tasks, namely Decision Tree (DT) [5,11,27], Support Vector Machine (SVM) [9, 17,21,23,25,26], and Artificial Neural Network (ANN) [9,29]. DT is used to predict building energy demand levels [27] and analyze the electricity load level based on hourly observations of the electricity load and weather [11]. Differently, Bansal et al. [5] use an evolved version of decision tree, Boosted Decision Tree Regression (BDTR), to model and forecast energy consumption so as to create personalized electricity plans for residential consumers based on usage history. The regression based on SVM is named Support Vector Regression (SVR). There are works using SVR to forecast power consumption [25] or using it in combination with other techniques, such as fuzzy-rough feature selection [23], particle swarm optimization algorithms [21], and chaotic artificial bee colony algorithm [17]. Gajowniczek and Zabkowski choose SVM and ANN because they believe that time-series analysis is not suitable in their work since they observe high volatility in the data [9]. Yu et al. [26] uses SVM and Backward Propagation Neural Network (BPNN), whose results show that SVM offers smaller prediction errors than BPNN. Zufferey et al. [29] apply Time Delay Neural Network (TDNN) and find out that the individual consumer’s consumption is harder to predict than an aggregation of multiple consumers. Very recently, Marino et al. [18] construct LSTM deep neural networks to forecast building energy load using historical consumption data. Despite There are three major machine-learning algorithms used in demand forecasting tasks, namely Decision Tree (DT) [5,11,27], Support Vector Machine (SVM) [9, 17,21,23,25,26], and Artificial Neural Network (ANN) [9,29]. DT is used to predict building energy demand levels [27] and analyze the electricity load level based on hourly observations of the electricity load and weather [11]. Differently, Bansal et al. [5] use an evolved version of decision tree, Boosted Decision Tree Regression (BDTR), to model and forecast energy consumption so as to create personalized electricity plans for residential consumers based on usage history. The regression based on SVM is named Support Vector Regression (SVR). There are works using SVR to forecast power consumption [25] or using it in combination with other techniques, such as fuzzy-rough feature selection [23], particle swarm optimization algorithms [21], and chaotic artificial bee colony algorithm [17]. Gajowniczek and Zabkowski choose SVM and ANN because they believe that time-series analysis is not suitable in their work since they observe high volatility in the data [9]. Yu et al. [26] uses SVM and Backward Propagation Neural Network (BPNN), whose results show that SVM offers smaller prediction errors than BPNN. Zufferey et al. [29] apply Time Delay Neural Network (TDNN) and find out that the individual consumer’s consumption is harder to predict than an aggregation of multiple consumers. Very recently, Marino et al. [18] construct LSTM deep neural networks to forecast building energy load using historical consumption data. Despite
A Comparative Analysis of Machine Learning Approaches for Short-/Long-Term Electricity Load Forecasting in Cyprus named paper contains the following information:
Power Demand Forecasting Using Long Short-Term Memory (LSTM) Deep-Learning Model for Monitoring Energy Sustainability named paper contains the following information: