Comparing ARIMA and LSTM models to predict time series in the oil industry

Monitoring and forecasting oil and gas (O\&G) production is essential to extend the life of a well and increase reservoirs' productivity. Popular models for O\&G time series are ARIMA and LSTM recurrent networks, and tipically several lags are forecasted at once. LSTM models can deploy the recursive prediction strategy, which uses one prediction to make the next, or the multiple outputs (MO) strategy, which predicts a sequence of values in a single shot. This work assesses ARIMA and LSTM models for the forecasting of petroleum production time series. We use time series of pressure and gas/oil flow from actual wells with distinct properties, for which we developed predictive models considering different time horizons. For the LSTM models, we deploy both the recursive and MO strategies. Our comparison revealed the superiority of LSTM models in general, and MO-based models for longer time intervals.

Prediction of Chlorophyll-a Concentrations in the Nakdong River Using Machine Learning Methods

Many studies have attempted to predict chlorophyll-a concentrations using multiple regression models and validating them with a hold-out technique. In this study commonly used machine learning models, such as Support Vector Regression, Bagging, Random Forest, Extreme Gradient Boosting (XGBoost), Recurrent Neural Network (RNN), and Long–Short-Term Memory (LSTM), are used to build a new model to predict chlorophyll-a concentrations in the Nakdong River, Korea. We employed 1–step ahead recursive prediction to reflect the characteristics of the time series data. In order to increase the prediction accuracy, the model construction was based on forward variable selection. The fitted models were validated by means of cumulative learning and rolling window learning, as opposed to the hold–out technique. The best results were obtained when the chlorophyll-a concentration was predicted by combining the RNN model with the rolling window learning method. The results suggest that the selection of explanatory variables and 1–step ahead recursive prediction in the machine learning model are important processes for improving its prediction performance.

BiHMP-GAN: Bidirectional 3D Human Motion Prediction GAN

Human motion prediction model has applications in various fields of computer vision. Without taking into account the inherent stochasticity in the prediction of future pose dynamics, such methods often converges to a deterministic undesired mean of multiple probable outcomes. Devoid of this, we propose a novel probabilistic generative approach called Bidirectional Human motion prediction GAN, or BiHMP-GAN. To be able to generate multiple probable human-pose sequences, conditioned on a given starting sequence, we introduce a random extrinsic factor r, drawn from a predefined prior distribution. Furthermore, to enforce a direct content loss on the predicted motion sequence and also to avoid mode-collapse, a novel bidirectional framework is incorporated by modifying the usual discriminator architecture. The discriminator is trained also to regress this extrinsic factor r, which is used alongside with the intrinsic factor (encoded starting pose sequence) to generate a particular pose sequence. To further regularize the training, we introduce a novel recursive prediction strategy. In spite of being in a probabilistic framework, the enhanced discriminator architecture allows predictions of an intermediate part of pose sequence to be used as a conditioning for prediction of the latter part of the sequence. The bidirectional setup also provides a new direction to evaluate the prediction quality against a given test sequence. For a fair assessment of BiHMP-GAN, we report performance of the generated motion sequence using (i) a critic model trained to discriminate between real and fake motion sequence, and (ii) an action classifier trained on real human motion dynamics. Outcomes of both qualitative and quantitative evaluations, on the probabilistic generations of the model, demonstrate the superiority of BiHMP-GAN over previously available methods.

ON-LINE MODELLING AND FORECASTING OF A MOVING CAR INFORMATION USING TD-HMLP NETWORK

The study proposed a model called trend data hybrid multilayered perceptron network (TD-HMLP) coupled with a modified recursive prediction error (MRPE) training algorithm as a nonlinear modeling. An on-line model was used to forecast speed, revolution and fuel balanced in a Proton Gen2 car tank. The car measured the injected fuel from fuel injection sensor and become an input for the TD-HMLP model to forecast the speed, revolution and fuel balanced in tank. These forecasted variables were also measured from the car sensors. The criterions for performances are based on the one step ahead forecasting (OSA), multi-step ahead forecasting (MSA) and adjusted R2. The forecasting result showed that TD-HMLP network is better than the conventional HMLP network to maintain higher value in adjusted R2 and produce better step in multi-step ahead forecasting. These preliminary results show that the proposed modeling approach is capable to be used as an on-line information forecaster of a moving car.