Predicting Prices in the Power TAC Wholesale Energy Market
Authors: Moinul Morshed Porag Chowdhury
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | I describe my work on using machine learning methods to predict prices in the Power TAC wholesale market, which will be used in future bidding strategies. I have been experimenting with three machine learning algorithms to predict prices: 1) REPTree (Decision tree) 2) Linear Regression and 3) Multilayer Perceptron (Neural network). I used mean absolute error as an evaluation metric for the performance of different versions of the prediction module. I have conducted experiments using Agent UDE15 (Agent UDE15 2015) as a variable agent and generated four test datasets using four different initializations. The experimental results shows that a decision tree model makes good predictions compared to other models. |
| Researcher Affiliation | Academia | Moinul Morshed Porag Chowdhury The University of Texas at El Paso El Paso, TX, USA mchowdhury4@miners.utep.edu |
| Pseudocode | No | The paper does not contain any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | No | The paper does not provide explicit statements about releasing source code for the described methodology or a direct link to a code repository. |
| Open Datasets | No | The paper describes generating training data: 'I ran one simulation for each of the five variable agent from the agent pool using four different initializations of the simulation. As a result, each initialization has five training datasets and in total I generated twenty training datasets.' However, it does not provide concrete access information (link, DOI, specific repository) to make these generated datasets publicly available. |
| Dataset Splits | No | The paper mentions '5, 10, 15 and 20 games training datasets' and 'four test datasets', but it does not specify any validation splits or percentages for the datasets used in the experiments. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU or CPU models, memory) used to run the experiments. |
| Software Dependencies | No | The paper mentions 'Weka (Hall et al. 2009 2015)' as a tool used, which provides a version range in its citation. However, it does not list other key software components with specific version numbers (e.g., programming languages, libraries, frameworks) required for reproducibility beyond Weka itself. |
| Experiment Setup | Yes | I have been experimenting with three machine learning algorithms to predict prices: 1) REPTree (Decision tree) 2) Linear Regression and 3) Multilayer Perceptron (Neural network). I chose some potential features that are available in the simulation at runtime to train a price predictor. I included 8 price features into my feature set because recent trading histories reflect the present wholesale market economy. These include the clearing prices for the previous hour, as well as the same time slot in the previous day and week. I also included the weather forecast data and time related data because the energy production of the producers (e.g. renewable energy sources as energy producers in the system) are related with this. I selected the number of participants in the game as a feature because the amount of competition affects the market clearing price. I also included the moving average prices predicted by the baseline SPOT agent as a feature so that I could get a better predictor model than the baseline. Each of the Linear regression, 3 layer Multilayer Perceptron (25,40,35 nodes respectively on each layer) and REPTree (Decision tree) predictor models has four different versions according to the number of games. |