With the strong capability of modeling time sequence, long short-term memory (LSTM) networks have been widely applied to predicting financial time series. This has attracted tremendous attention in the quantitative trading area. A complete quantitative trading system usually has three tasks, including market timing, stock selection, and portfolio management. In this paper, we present an LSTM-based quantitative trading system and optimize this system from the following two aspects. Firstly, in the process of stock selection, we first introduce the dynamic K-top method in the LSTM-based quantitative trading system to follow the market change. Secondly, concerning portfolio management, we further incorporate the Kelly Criterion to attain an appropriate position ratio. Taking CSI300 constituent stocks as the study example, extensive experiments have been carried out to show the superiority of the proposed method. In comparison with the straight forward LSTM-based trading strategy, the improved LSTM-based trading strategy with the dynamic K-top method and the Kelly Criterion can achieve an increase of 44.97% over ten days in terms of accumulative return. In addition, our novel method can gain a win ratio of 55.95%, a monthly alpha of 0.16, a monthly Sharpe ratio of 2.17, and a monthly Sortino ratio of 2.96 disregarding the transaction costs.

Wind speed prediction and forecasting is important for various business and management sectors. In this paper, we introduce new models for wind speed prediction based on graph convolutional networks (GCNs). Given hourly data of several weather variables acquired from multiple weather stations, wind speed values are predicted for multiple time steps ahead. In particular, the weather stations are treated as nodes of a graph whose associated adjacency matrix is learnable. In this way, the network learns the graph spatial structure and determines the strength of relations between the weather stations based on the historical weather data. We add a self-loop connection to the learnt adjacency matrix and normalize the adjacency matrix. We examine two scenarios with the self-loop connection setting (two separate models). In the first scenario, the self-loop connection is imposed as a constant additive. In the second scenario a learnable parameter is included to enable the network to decide about the self-loop connection strength. Furthermore, we incorporate data from multiple time steps with temporal convolution, which together with spatial graph convolution constitutes spatio-temporal graph convolution. We perform experiments on real datasets collected from weather stations located in cities in Denmark and the Netherlands. The numerical experiments show that our proposed models outperform previously developed baseline models on the referenced datasets. We provide additional insights by visualizing learnt adjacency matrices from each layer of our models.

Wind speed forecasting has proven to be crucial for many domains like power generation. However, it is quite convoluted and demanding due to the unpredictability of the wind. Numerous physical and statistical methods have been employed in the past by the researchers in India and over the years, these methods have been improvised. While the physical methods prove to be computationally expensive and the statistical methods represent weaker generalizations of non-linear features, there arises a need to develop much more efficient and robust approaches that can augment with the traditional methodologies. Artificial Intelligence has shown tremendous potential in providing solutions to time-series forecasting tasks, akin to the precise task that this work aims to address. Various machine learning and deep learning algorithms have been found to be a prominent choice for forecasting of various weather parameters. This work aims at implementing two machine learning algorithms, namely, Light Gradient Boosting Machine (LGBM) and Long Short-Term Memory (LSTM) network to obtain short-term forecasts of wind speed of Indian weather stations. Robust feature engineering approaches like Miss Forest to fill in the missing values, a novel feature representation for giving importance to recent data, and Fast Fourier Transform (FFT) with Digital Filters for removal of outliers and noises in the data have been used. Finally, stacking, an ensemble learning has been implemented using LGBM and LSTM as base learners and Random Forest as the metalearner. The average Root Mean Square Error (RMSE) (in m/s to tenths) obtained by individual models is high. 24-hour wind forecast errors of LGBM and LSTM are 8.19 and 21.17, whereas the forecast error using the proposed ensemble framework is 5.74.

Wind energy, one of the greatest progressing renewable energy sources, becomes more significant for sustainable development and environmental protection. Its intermittent nature makes accurate and ...

We propose a time series forecasting method for the future prices of agricultural products and present the criteria by which forecasted future time series are evaluated in the context of statistical characteristics. Time series forecasting of agricultural products has the basic importance in maintaining the sustainability of agricultural production. The prices of agricultural products show seasonality in their time series, and conventional methods such as the auto-regressive integrated moving average (ARIMA or the Box Jenkins method) have tried to exploit this feature for forecasting. We expect that recurrent neural networks, representing the latest machine learning technology, can forecast future time series better than conventional methods. The measures used in evaluating the forecasted results are also of importance. In literature, the accuracy determined by the error rate at a specific time point in the future, is widely used for evaluation. We predict that, in addition to the error rate, the criterion for conservation of the statistical characteristics of the probability distribution function from the original past time series to the future time series in the forecasted future is also important. This is because some time series have a non-Gaussian probability distribution (such as the Levy stable distribution) as a characteristic of the target system; for example, market prices on typical days change slightly, however on certain occasions, change dramatically. We implemented two methods for time series forecasting based on recurrent neutral network (RNN), one of which is called time-alignment of time point forecast (TATP), and another one is called direct future time series forecast (DFTS). They were evaluated using the two aforementioned criteria consisting of the accuracy and the conservation of the statistical characteristics of the probability distribution function. We found that after intensive training, TATP of LTSM shows superior performance in not only accuracy, but also the conservation compared to TATP of other RNNs. In DFTS, DFTS of LSTM cannot show the best performance in accuracy in RMS sense, but it shows superior performance in other criteria. The results suggest that the selection of forecasting methods depends on the evaluation criteria and that combinations of forecasting methods is useful based on the application. The advantage of our method is that the required length of time series for training is enough short, namely, we can forecast the whole cycle of future time series after training with even less than the half of the cycle, and it can be applied to the field where enough numbers of continuous data are not available.

We consider predicting systemic financial crises one to five years ahead using recurrent neural networks. The prediction performance is evaluated with the Jorda-Schularick-Taylor dataset, which includes the crisis dates and relevant macroeconomic series of 17 countries over the period 1870-2016. Previous literature has found simple neural network architectures to be useful in predicting systemic financial crises. We show that such predictions can be greatly improved by making use of recurrent neural network architectures, especially suited for dealing with time series input. The results remain robust after extensive sensitivity analysis.

Water scarcity is a burgeoning problem of our society. Population expansion and the proliferation of apartments in cities have put tremendous pressure on potable water. Smart water meters that monitor consumption, have been increasingly deployed to encourage people against consuming excessive water, by making them aware of their consumption on a daily to monthly basis. However, the efforts in achieving the desired outcomes of conservation have been unsatisfactory, largely due to the negligence of the consumer and the inability of the systems to pre-empt such attitudes. This paper describes a novel system that corroborates positive attitudes towards water conservation. The proposed system consists of a smart card based, intelligent water meter that is based on IoT and uses a differential consumer pricing policy and an embedded seasonal auto regressive integrated moving average (SARIMA) model to forecast water bills at the edge.

Variables local to methods play an important role in implementing not only the logic of desired algorithms but also the control of global variables. The values of these variables dictate the paths to be taken while executing the program. While testing is an effective way to exercise the programs under test, the values of these local variables determines the reachability of execution paths. Hence, it is possible that these local variables control the execution of the program and thus may result in some paths never to be tested adequately. This paper introduces a methodology and a prototype tool, called "TestLocal", to enable testing of methods by changing the values of local variables at runtime thus enabling traversing the paths that are hard to execute. By changing the values of local variables during the execution, the unexpected paths are intentionally forced to be executed. In comparison with the existing approaches (e.g., Microsoft IntelliTest), the proposed methodology requires little to no modifications to the given code. The novelty of the proposed approach is that it integrates a constraint solver and a debugger to forcefully execute unreachable paths.

Traffic Congestion wastes time and energy, which are the two most valuable commodities of the current century. It happens when too many vehicles try to use a transportation infrastructure without having enough capacity. However, researches indicate that adding extra lane without studying the future consequences does not improve the situation. Our goal is to add another layer of information to the traffic data, find which type of vehicles are contributing to road traffic congestion, and predict future road traffic congestion and demands based on the historical data. We collected more than 400,000 images from traffic cameras installed in Autoroute 40, in the city of Montreal. The images were collected for five consecutive weeks from different locations from April 14, 2019, up until May 18, 2019. To process these images and extract useful information out of them, we created an object detection and classification model using the Faster RCNN algorithm. Our goal was to be able to detect different types of vehicles and see if we have traffic congestion in an image. In order to improve the accuracy and reduce the error rate, we provided multiple examples with different conditions to the model. By introducing blurry, rainy, and low light images to the model, we managed to build a robust model that could do the detection and classification task with excellent accuracy. Furthermore, by extracting the information from the collected images, we created a dataset of the number of vehicles in each location. After analyzing and visualizing the data, we find out the most congested areas, the behavior of the traffic flow during the day, peak hours, the contribution of each type of vehicle to the traffic, seasonality of the data, and where we can see each type of vehicle the most. Finally, we managed to predict the total number of congestion incidents for seven days based on historical data. Besides, we were able to predict the total number of different types of vehicles on the road as well. In order to do this task, we developed multiple Regression, Deep Learning, and Time Series Forecasting models and trained them with our vehicle count dataset. Based on the experimental results, we were able to get the best predictions with the Deep Learning models and succeeded in predicting future road traffic congestion with excellent accuracy.

To solve the problem of predicting daily arrival number of people belonging to a specific group, this paper proposes a hybrid model which integrates linear component (from ARIMA model) with non-linear component (from BP neural network). The experimental results indicate that, hybrid methodology that has both linear and nonlinear modeling capabilities is a good strategy for practical use.

To ensure effective management and security in large-scale public events, it is imperative for the event organizers to be aware of potentially critical crowd densities. This article, therefore, presents a solution to the above problem in terms of WiFi-based crowd counting and long short-term memory (LSTM) neural network-based forecasting. Monitoring of an actual event organized in Brussels has been described, wherein crowd counts are obtained using WiFi sensors in a privacy-preserved manner. The time-stamped crowd counts are used to develop univariate time-series, which are in-turn utilized for forecasting. Five different LSTM models are utilized for crowd time-series forecasting and analyzed for their suitability. A random walk model is used as a reference for performance assessment. Among different LSTM models, Convolutional LSTM delivered the best performance. Overall results and analysis show that the developed system is suitable for crowd monitoring.

Time series forecasting predicts values in future timestamps based on historically observed series information. Algorithms based on deep neural networks such as Temporal Convolution Network(TCN) have outperformed traditional methods such as Autoregressive Integrated Moving Average Model(ARIMA). However, most existing deep learning approaches suffer from the insufficient ability to capture the seasonality features in series adequately since the network structure ignores the fact that the importance of points the series varies a lot. The local context that reflects a sub-segment of seasonality can indicate potential patterns of the series based on the periodicity. Therefore, we tend to exploit local information from historical records. To this end, we develop a novel strategy to extract local context sensitivity information and integrate them into the current state-of-the-art TCN model, namely LS-TCN. This information enables an improvement in capturing the series pattern and fluctuation, as well as providing transferable guidance for forecasting in the next steps. Experiments conducted on three different real-world series datasets demonstrate that our method significantly outperforms the state-of-the-art models, especially in autocorrelation series corpus.

Time series forecasting has gained lots of attention recently; this is because many real-world phenomena can be modeled as time series. The massive volume of data and recent advancements in the processing power of the computers enable researchers to develop more sophisticated machine learning algorithms such as neural networks to forecast the time series data. In this paper, we propose various neural network architectures to forecast the time series data using the dynamic measurements; moreover, we introduce various architectures on how to combine static and dynamic measurements for forecasting. We also investigate the importance of performing techniques such as anomaly detection and clustering on forecasting accuracy. Our results indicate that clustering can improve the overall prediction time as well as improve the forecasting performance of the neural network. Furthermore, we show that feature-based clustering can outperform the distance-based clustering in terms of speed and efficiency. Finally, our results indicate that adding more predictors to forecast the target variable will not necessarily improve the forecasting accuracy.

Time series data refers to data points that follow a chronological sequence or ordering. Modelling and analysis of such data is generally done to extract significant statistics and also to utilize the past historic data in order to predict future points. In this paper, three popular time series forecasting methods, namely Holt Winters, ARIMA and LSTMs, are applied on CPU data and their results are compared. LSTM is found to be more suited for predicting CPU usage followed by ARIMA. LSTM performs better due to the fact that CPU usage is unstable and has fluctuations even though it is seasonal in nature. By performing such an analysis, it is possible to identify patterns which help in predicting the usage of future resources for future demand which in turn enables optimization of resource management.

This technical quest aspired to build deep multifaceted system proficient in forecasting banana harvest yields essential for extensive planning for a sustainable production in the agriculture sector. Recently, deep-learning (DL) approach has been used as a new alternative model in forecasting. In this paper, the enhanced DL approach incorporates multiple long short term memory (LSTM) layers employed with multiple neurons in each layer, fully trained and built a state for forecasting. The enhanced model used the banana harvest yield data from agrarian reform beneficiary (ARB) cooperative of Dapco in Davao del Norte, Philippines. The model parameters such as epoch, batch size and neurons underwent tuning to identify its optimal values to be used in the experiments. Additionally, the root-mean-squared error (RMSE) is used to evaluate the performance of the model. Using the same set of training and testing data, experiment exhibits that the enhanced model achieved the optimal result of 34.805 in terms of RMSE. This means that the enhanced model outperforms the single and multiple LSTM layer with 43.5 percent and 44.95 percent reduction in error rates, respectively. Since there is no proof that LSTM recurrent neutral network has been used with the same agricultural problem domain, therefore, there is no standard available with regards to the level of error reduction in the forecast. Moreover, investigating the performance of the model using diverse datasets specifically with multiple input features (multivariate) is suggested for exploration. Furthermore, extending and embedding this approach to a web-based along with a handy application is the future plan for the benefit of the medium scale banana growers of the region for efficient and effective decision making and advance planning.

This paper presents a process of obtaining a forecast of the demand for electricity in the Empresa Eléctrica Ambato Regional Centro Norte S.A. (EEASA) using Big Data criteria. A computational model in programming language PYTHON 3.7 was developed. Implementing an electric load projection application in which machine learning algorithms are integrated, such as Autoregressive Integrated Moving Average, Linear Regression, Support Vector Machine and the Multilayer Perceptron. Additionally, different study cases were considered as the most frequent scenarios for the generation of electric load projections such as: a week with holidays, a week with scheduled and unscheduled incidence, a week in dry season and week in rainy season, which allowed test each implemented algorithm, evaluate its results and compare them with each other by calculating the indicators of the percentage of average absolute error (PEMA) and the accuracy of the projection (EP).

This paper introduces an approach based on control theory to model, analyze and select optimal security policies for Moving Target Defense (MTD) deployment strategies. A Markov Decision Process (MDP) scheme is presented to model states of the system from attacking point of view. The employed value iteration method is based on the Bellman optimality equation for optimal policy selection for each state defined in the system. The model is then utilized to analyze the impact of various costs on the optimal policy. The MDP model is then applied to two case studies to evaluate the performance of the model.

This paper introduces a two-stage deep learning-based methodology for clustering time series data. First, a novel technique is introduced to utilize the characteristics (e.g., volatility) of the given time series data in order to create labels and thus enable transformation of the problem from an unsupervised into a supervised learning. Second, an autoencoder-based deep learning model is built to model both known and hidden non-linear features of time series data. The paper reports a case study in which the selected financial and stock time series data of over 70 stock indices are clustered into distinct groups using the introduced two-stage procedure. The results show that the proposed methodology is capable of achieving 87.5% accuracy in clustering and predicting the labels for unseen time series data. The paper also reports an important finding in which it is observed that the performance of both techniques (i.e., autoencoder and Kmeans) are comparable. However, there are a few instances of time series data that are classified differently by the autoencoder-based methodology compared to the Kmeans algorithm. The results may indicate that the proposed deep learning-based approach is taking into account additional hidden features that might be overlooked by conventional Kmeans. The finding raises the question whether the explicit features of data should be analyzed for clustering or more advanced techniques such as deep learning need to be adapted by which hidden features and relationships are explored for clustering purposes.

This paper applies a recurrent neural network (RNN) method to forecast cotton and oil prices. We show how these new tools from machine learning, particularly Long-Short Term Memory (LSTM) models, complement traditional methods. Our results show that machine learning methods fit reasonably well the data but do not outperform systematically classical methods such as Autoregressive Integrated Moving Average (ARIMA) models in terms of out of sample forecasts. However, averaging the forecasts from the two type of models provide better results compared to either method. Compared to the ARIMA and the LSTM, the Root Mean Squared Error (RMSE) of the average forecast was 0.21 and 21.49 percent lower respectively for Cotton. For Oil the forecast averaging does not provide improvements in terms of RMSE. We suggest using a forecast averaging method and extending our analysis to a wide range of commodity prices.

The technology used in the printing industry is currently growing rapidly. Generally, the digital printing industry uses raw materials in the form of paper production. The use of paper material with large volumes is clear badly in need of purchasing large quantities of paper stock as well. The purchase of paper stocks with a constant amount at the beginning of each month for various types of paper causes a buildup or lack of material stock standard on certain types of paper. During this time the purchase and ordering of raw materials only based on the estimates or predictions of the owner. In this paper proposed forecasting will be carried out in the digital printing industry by applying the ARIMA model for each type of raw material paper with the Palembang F18 digital printing case study. The ARIMA modeling applied will produce different parameters for each materials paper type so as to produce forecasting with the Akaike Information Criterion (AIC) value averages 13.0294%.