Mitigating the Weaknesses of Machine Learning in Short–Term Forecasting of Aggregated Power System Active Loads

Koponen, Pekka; Niska, Harri; Mutanen, Antti

doi:10.1109/indin41052.2019.8972182

Cited by 2 publications

(6 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The results suggest that improving the short-term forecasting accuracy of the aggregated active demand loads further from our best recent methods [11] now may not give adequately significant savings in the expected value of the imbalance costs to justify much method development investments. It seems more important to mitigate the risks caused by high imbalance price peaks.…”

Section: Discussionmentioning

confidence: 98%

“…It summarizes some results from our past publications between the years 2012 and 2020. All these publications can be found via [11,16]. All the most accurate methods in Figure 3 are hybrids that combine several short-term forecasting methods and include both machine learning and physically based models.…”

Section: Comparison Of Methods Across Different Forecasting Casesmentioning

confidence: 99%

“…When studying and developing short term load forecasting methods for active demand, such as [11], we have detected the following challenges:…”

Section: Needs To Develop Short Term Load Forecasting Criteriamentioning

confidence: 99%

“…For some methods, this is a relevant problem also in off-line model identification, especially because the models need to be updated due to changes in the grid, market aggregation, customer behavior, new priorities, etc. For example, support vector regression (SVR) has so poor computational scalability that it limits the possibilities to exploit it in 1 or 3 min time resolution short-term forecasting needed in electricity grid operation [11]. Based on a brief discussion and comparison of computation times of their models in short term forecasting, Lusis et al [4] concluded that model run time in model training may be an important factor when choosing between models.…”

Section: Computational Timementioning

confidence: 99%

“…Physically based model structures with power constraints can also be used to model the main phenomenon that contributes to the peak load. We have improved forecasting of rarely occurring and extreme situations by combining several different models into hybrid models, see [11] for an example. In order to better assess, compare and develop methods, we need forecasting criteria that adequately reflect the concentration of the economic costs of the forecasting errors to the high load situations.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Assessing and Comparing Short Term Load Forecasting Performance

et al. 2020

Self Cite

View full text Add to dashboard Cite

When identifying and comparing forecasting models, there may be a risk that poorly selected criteria could lead to wrong conclusions. Thus, it is important to know how sensitive the results are to the selection of criteria. This contribution aims to study the sensitivity of the identification and comparison results to the choice of criteria. It compares typically applied criteria for tuning and performance assessment of load forecasting methods with estimated costs caused by the forecasting errors. The focus is on short-term forecasting of the loads of energy systems. The estimated costs comprise electricity market costs and network costs. We estimate the electricity market costs by assuming that the forecasting errors cause balancing errors and consequently balancing costs to the market actors. The forecasting errors cause network costs by overloading network components thus increasing losses and reducing the component lifetime or alternatively increase operational margins to avoid those overloads. The lifetime loss of insulators, and thus also the components, is caused by heating according to the law of Arrhenius. We also study consumer costs. The results support the assumption that there is a need to develop and use additional and case-specific performance criteria for electricity load forecasting.

show abstract

Section: Discussionmentioning

confidence: 98%

Section: Comparison Of Methods Across Different Forecasting Casesmentioning

confidence: 99%

“…When studying and developing short term load forecasting methods for active demand, such as [11], we have detected the following challenges:…”

Section: Needs To Develop Short Term Load Forecasting Criteriamentioning

confidence: 99%

Section: Computational Timementioning

confidence: 99%

mentioning

confidence: 99%

See 3 more Smart Citations

Assessing and Comparing Short Term Load Forecasting Performance

et al. 2020

Self Cite

View full text Add to dashboard Cite

show abstract

Building Heat Demand Forecasting by Training a Common Machine Learning Model with Physics-Based Simulator

et al. 2021

View full text Add to dashboard Cite

Accurate short-term forecasts of building energy consumption are necessary for profitable demand response. Short-term forecasting methods can be roughly classified into physics-based modelling and data-based modelling. Both of these approaches have their advantages and disadvantages and it would be therefore ideal to combine them. This paper proposes a novel approach that allows us to combine the best parts of physics-based modelling and machine learning while avoiding many of their drawbacks. A key idea in the approach is to provide a variety of building parameters as input for an Artificial Neural Network (ANN) and train the model with data from a large group of simulated buildings. The hypothesis is that this forces the ANN model to learn the underlying simulation model-based physics, and thus enables the ANN model to be used in place of the simulator. The advantages of this type of model is the combination of robustness and accuracy from a high-detail physics-based model with the inference speed, ease of deployment, and support for gradient based optimization provided by the ANN model. To evaluate the approach, an ANN model was developed and trained with simulated data from 900–11,700 buildings, including equal distribution of office buildings, apartment buildings, and detached houses. The performance of the ANN model was evaluated with a test set consisting of 60 buildings (20 buildings for each category). The normalized root mean square errors (NRMSE) were on average 0.050, 0.026, 0.052 for apartment buildings, office buildings, and detached houses, respectively. The results show that the model was able to approximate the simulator with good accuracy also outside of the training data distribution and generalize to new buildings in new geographical locations without any building specific heat demand data.

show abstract

Mitigating the Weaknesses of Machine Learning in Short–Term Forecasting of Aggregated Power System Active Loads

Cited by 2 publications

References 19 publications

Assessing and Comparing Short Term Load Forecasting Performance

Assessing and Comparing Short Term Load Forecasting Performance

Building Heat Demand Forecasting by Training a Common Machine Learning Model with Physics-Based Simulator

Contact Info

Product

Resources

About