Forecasting residential gas consumption with machine learning algorithms on weather data

Brian, de Keijzer,; Visser, Pol de; Romillo, Víctor García; Muñoz, Víctor Gómez; Boesten, Daan; Meezen, Megan; Rahola, T.B. Salcedo

doi:10.1051/e3sconf/201911105019

“…After removing these outliers, the remaining population help to stabilize the model performance and speed the convergence progress. Table III shows the comparison of the proposed method and its variations with the CNN model in [8]. The results show that the proposed method can achieve reasonable accuracy level with a simple architecture for both single-output model and the multi-output parallel process model.…”

Section: A Performancementioning

confidence: 99%

“…An accurate prediction of the gas usage can help the supplier to match the supply with demand and operate the system more efficiently. Several attempts have been made to feed the data collected by smart meters into pre-trained models to predict the domestic consumption of natural gas [5]- [8]. However, most of the studies produce predictions of the overall gas consumption using data collected by smart meters.…”

mentioning

confidence: 99%

Transformation based Parallel Forecasting of Individual Gas Consumption from Smart Meter Data

Hu¹,

Sikdar²

2023

Preprint

0

View full text Add to dashboard Cite

<p>The prediction of energy consumption of users is an important requirement for utilities. Machine learning using neural networks has been applied for predicting household energy consumption due to effectiveness in capturing correlations and patterns. However, the result is predicted by learning over the behavior of the training data as a whole and ignoring the differences between sub-group behavior (i.e., groups of users with similar characteristics). The prediction of energy consumption of users is an important requirement for utilities. Exploiting the behavioral differences between different subgroups in the population can improve the system’s capability to perform parallel processing of similar tasks. This paper proposes an innovative model to simultaneously predict gas consumption of multiple households or population subgroups. The proposed method is based on the use of data transformations and focuses on differences in individual behavior. The evaluation of the model is discussed from three aspects: prediction accuracy, risk of privacy, as well as system robustness. The results shows that the proposed system can achieve multiple prediction simultaneously with reasonable accuracy and be secure against noise and attacks on private information. </p>

show abstract

“…The model is simple in structure and the study demonstrates the feasibility of applying neural networks for natural gas consumption prediction although there are pre-processing and optimization requirements for input data prior to their use in the neural network. An improved version of neural network model [8] was proposed by taking weather condition into consideration to obtain a good consumption prediction even without a series of meter data.…”

Section: Related Workmentioning

confidence: 99%

Transformation based Parallel Forecasting of Individual Gas Consumption from Smart Meter Data

Hu¹,

Sikdar²

2023

Preprint

0

View full text Add to dashboard Cite

<p>The prediction of energy consumption of users is an important requirement for utilities. Machine learning using neural networks has been applied for predicting household energy consumption due to effectiveness in capturing correlations and patterns. However, the result is predicted by learning over the behavior of the training data as a whole and ignoring the differences between sub-group behavior (i.e., groups of users with similar characteristics). The prediction of energy consumption of users is an important requirement for utilities. Exploiting the behavioral differences between different subgroups in the population can improve the system’s capability to perform parallel processing of similar tasks. This paper proposes an innovative model to simultaneously predict gas consumption of multiple households or population subgroups. The proposed method is based on the use of data transformations and focuses on differences in individual behavior. The evaluation of the model is discussed from three aspects: prediction accuracy, risk of privacy, as well as system robustness. The results shows that the proposed system can achieve multiple prediction simultaneously with reasonable accuracy and be secure against noise and attacks on private information. </p>

show abstract

“…The best forecast methods are random forest and regression tree approaches, depending on the weather data applied. Finally, de Keijzer, de Visser, Romillo, Muñoz, Boesten, Meezen, and Rahola (2019) forecast the natural gas usage of 52 homes in the Netherlands using six methods applying hourly data. The most accurate hourly forecast was the deep neural network, and multivariate linear regression was applied daily.…”

Section: Gas Consumption Forecastingmentioning

confidence: 99%

Long-Term Optimal Delivery Planning for Replacing the Liquefied Petroleum Gas Cylinder

Yoshida¹,

Sato²,

Uchiumi³

et al. 2021

Preprint

0

View full text Add to dashboard Cite

This study proposes a method for efficient delivery of liquefied petroleum gas cylinders based on demand forecasts of gas usage. To maintain a liquefied petroleum gas service, gas providers visit each customer to check the gas meter and replace the gas cylinder depending on the remaining amount of gas. These visits can be categorized into three patterns: non-replacement visit, replacement before the customer runs out of gas, and replacement after the customer runs out of gas. The last pattern is a crucial problem in sustaining a liquefied petroleum gas service, and it must be reduced. By contrast, frequent non-replacement visits are required to prevent the customer from running out of gas, but it requires considerable effort. One of the most severe difficulties of this problem is acquiring the gas usages of each customer only by visiting. However, with the recent spread of smart sensors, the daily gas consumption of each house can be monitored without having to visit customers. Such sensors can be utilized to solve the above problems. Our main idea is to categorize all customers into three groups: high-risk, moderate-risk, and lowrisk by focusing on an urgent need for cylinder replacement based on the demand forecast. Based on this idea, we construct an algorithm to maximize the delivery for moderate-risk customers while ensuring delivery to high-risk customers. Long-term optimal delivery planning is realized by achieving workload balancing per day. The verification experiment in Chiba prefecture in Japan showed the effectiveness of our algorithm in reducing the number of out-of-gas cylinders. Moreover, the proposed model is a new generic framework for building an optimal vehicle routing plan, consisting of a complementary algorithm, demand forecast, delivery list optimization, and delivery route optimization for realizing a long-term optimal delivery plan by setting the priority. The delivery optimization of other areas such as the grocery and oil industries incurs similar problems. Therefore, our model has significant potential for solving real-world delivery problems.

show abstract

Forecasting residential gas consumption with machine learning algorithms on weather data

Cited by 10 publications

References 15 publications

Transformation based Parallel Forecasting of Individual Gas Consumption from Smart Meter Data

Transformation based Parallel Forecasting of Individual Gas Consumption from Smart Meter Data

Transformation based Parallel Forecasting of Individual Gas Consumption from Smart Meter Data

Long-Term Optimal Delivery Planning for Replacing the Liquefied Petroleum Gas Cylinder

Contact Info

Product

Resources

About