Forecasting Strategies Research Articles

Quantitative Trading Strategy Based on Simplified DPG

In recent years, investment has become more and more popular, and asset management has also received more and more attention. At the same time, with the development of computer science, more and more machine learning or deep learning algorithms can be used for investment management, such as price forecasting, portfolio and quantitative trading strategies. First of all, in the trading market, most of the ups and downs are cyclical, but they are easily affected by factors such as policies and investment fever, which brings great challenges to the establishment of a reasonable price forecasting model. This paper builds a sliding price forecasting model that aims to reduce various noises such as frequent fluctuations. In order to get the best forecast results, we searched the parameters of each sliding forecast, trying to get the forecast closest to the actual price. Second, developing trading strategies based on forecast results is clearly related to goal planning. Therefore, this paper first establishes the objective programming problem, designs the objective function and constraints, and then optimizes the problem. At the same time, the article also considers investor styles, such as the investment boom in gold and bitcoin, and a more cautious approach to buying and selling. Ultimately, this paper designs a flexible feedback model based on style parameters to adjust the trading strategy, by searching for the best parameters of the final model, our trading strategy model converts assets from $1000 from September 11, 2016 to September 10, 2021 It increased to $238,869, and the cumulative return reached 23,886.9%.

Predicting CO2 Emissions from Traffic Vehicles for Sustainable and Smart Environment Using a Deep Learning Model

Burning fossil fuels results in emissions of carbon dioxide (CO2), which significantly contributes to atmospheric changes and climate disturbances. Consequently, people are becoming concerned about the state of the environment, and governments are required to produce precise projections to develop efficient preventive measures. This study makes a significant contribution to the area by modeling and predicting the CO2 emissions of vehicles using advanced artificial intelligence. The model was constructed using the CO2 emission by vehicles dataset from Kaggle, which includes different parameters, namely, vehicle class, engine size (L), cylinder transmission, fuel type, fuel consumption city (L/100 km), fuel consumption hwy (L/100 km), fuel consumption comb (L/100 km), fuel consumption comb (mpg), and CO2 emissions (g/km). To forecast the CO2 emissions produced by vehicles, a deep learning long short-term memory network (LSTM) model and a bidirectional LSTM (BiLSTM) model were developed. Both models are efficient. Throughout the course of the investigation, the researchers employed four statistical assessment metrics: the mean square error (MSE), the root MSE (RMSE), Pearson’s correlation coefficient (R%), and the determination coefficient (R2). Based on the datasets of experiments carried out by Kaggle, the LSTM and BiLSTM models were created and implemented. The data were arbitrarily split into two phases: training, which included 80% of the total data, and testing, which comprised 20% of the total data. The BiLSTM model performed best in terms of accuracy and achieved high prediction values for MSE and RMSE. The BiLSTM model has the greatest attainable (R2 = 93.78). In addition, R% was used to locate a connection between the dataset’s characteristics to ascertain which characteristics had the highest level of association with CO2 emissions. An original strategy for the accurate forecasting of carbon emissions was developed as a result of this work. Consequently, policymakers may use this work as a potentially beneficial decision-support tool to create and execute successful environmental policies.

Solar Wind Data Assimilation in an Operational Context: Use of Near‐Real‐Time Data and the Forecast Value of an L5 Monitor

AbstractFor accurate and timely space weather forecasting, advanced knowledge of the ambient solar wind is required, both for its direct impact on the magnetosphere and for accurately forecasting the propagation of coronal mass ejections to Earth. Data assimilation (DA) combines model output and observations to form an optimum estimation of reality. Initial experiments with assimilation of in situ solar wind speed observations suggest the potential for significant improvement in the forecast skill of near‐Earth solar wind conditions. However, these experiments have assimilated science‐quality observations, rather than near‐real‐time (NRT) data that would be available to an operational forecast scheme. Here, we assimilate both NRT and science observations from the Solar Terrestrial Relations Observatory (STEREO) and near‐Earth observations from the Advanced Composition Explorer and Deep Space Climate Observatory spacecraft. We show that solar wind speed forecasts using NRT data are comparable to those based on science‐level data. This suggests that an operational solar wind DA scheme would provide significant forecast improvement, with reduction in the mean absolute error of solar wind speed around 46% over forecasts without DA. With a proposed space weather monitor planned for the L5 Lagrange point, we also quantify the solar wind forecast gain expected from L5 observations alongside existing observations from L1. This is achieved using configurations of the STEREO and L1 spacecraft. There is a 15% improvement for forecast lead times of less than 5 days when observations from L5 are assimilated alongside those from L1, compared to assimilation of L1 observations alone.

Forecasting Equatorial Ionospheric Convective Instability With ICON Satellite Measurements

AbstractMeasurements from the Ionospheric Connections Explorer satellite (ICON) form the basis of direct numerical forecast simulations of plasma convective instability in the postsunset equatorial F region ionosphere. ICON data are selected and used to initialize and force the simulations and then to test the results one orbit later when the satellite revisits the same longitude. Data from the IVM plasma density and drifts instrument and the MIGHTI red‐line thermospheric winds instrument are used to force the simulation. Data from IVM are also used to test for irregularities (electrically polarized plasma depletions). Fourteen datasets from late March 2022, were examined. The simulations correctly predicted the occurrence or non‐occurrence of irregularities 12 times while producing one false positive and one false negative. This demonstrates that the important telltales of instability are present in the ICON state variables and that the important mechanisms for irregularity formation are captured by the simulation code. Possible refinements to the forecast strategy are discussed.

Flash Floods: Forecasting, Monitoring and Mitigation Strategies

In recent decades, flash floods have become a major natural disaster and show a continuously increasing trend on a worldwide scale [...]

Prospect Forecast and Rectification Strategy of American Takeout Industry

The global pandemic has a serious impact on economic and social development of all countries. The production and demand of many industries have greatly shrunk. However, the takeout industry in the United States is an exception. Under the influence of COVID-19, the supply and demand of American takeout industry have increased significantly compared with that before the pandemic. The market has become increasingly hot. Once the current market environment changes, it is uncertain whether the American takeout industry can continue to be popular. Due to this reality, the research explores the market trend of American takeout industry in recent years, and forecasts the future takeout market. The study also analyzes whether the American takeout market should follow the Chinese market’s approach. The research reveals that there are several disadvantages in the American takeout market in terms of labor cost and the price of buyers. To address these issues, the study develops new sale method through several ways, for example, reducing the buyer's cost of delivery fee, making the delivery fee paid by the seller as well as by the platform. Moreover, the platform can send vouchers to the buyers together with seller.

Prediction Error-Based Power Forecasting of Wind Energy System Using Hybrid WT–ROPSO–NARMAX Model

The volatility and intermittency of wind energy result in highly unpredictable wind power output, which poses challenges to the stability of the intact power system when integrating large-scale wind power. The accuracy of wind power prediction is critical for maximizing the utilization of wind energy, improving the quality of power supply, and maintaining the stable operation of the power grid. To address this challenge, this paper proposes a novel hybrid forecasting model, referred to as Hybrid WT–PSO–NARMAX, which combines wavelet transform, randomness operator-based particle swarm optimization (ROPSO), and non-linear autoregressive moving average with external inputs (NARMAX). The model is specifically designed for power generation forecasting in wind energy systems, and it incorporates the interactions between the wind system’s supervisory control and data acquisition’s (SCADA) actual power record and numerical weather prediction (NWP) meteorological data for one year. In the proposed model, wavelet transform is utilized to significantly improve the quality of the chaotic meteorological and SCADA data. The NARMAX techniques are used to map the non-linear relationship between the NWP meteorological variables and SCADA wind power. ROPSO is then employed to optimize the parameters of NARMAX to achieve higher forecasting accuracy. The performance of the proposed model is compared with other forecasting strategies, and it outperforms in terms of forecasting accuracy improvement. Additionally, the proposed Prediction Error-Based Power Forecasting (PEBF) approach is introduced, which retrains the model to update the results whenever the difference between forecasted and actual wind powers exceeds a certain limit. The efficiency of the developed scheme is evaluated through a real case study involving a 180 MW grid-connected wind energy system located in Shenyang, China. The proposed model’s forecasting accuracy is evaluated using various assessment metrics, including mean absolute error (MAE) and root mean square error (RMSE), with the average values of MAE and RMSE being 0.27% and 0.30%, respectively. The simulation and numerical results demonstrated that the proposed model accurately predicts wind output power.

Assessing and forecasting water quality in the Danube River by using neural network approaches

The health and quality of the Danube River ecosystems is strongly affected by the nutrients loads (N and P), degree of contamination with hazardous substances or with oxygen depleting substances, microbiological contamination and changes in river flow patterns and sediment transport regimes. Water quality index (WQI) is an important dynamic attribute in the characterization of the Danube River ecosystems health and quality. The WQ index scores do not reflect the actual condition of water quality. We proposed a new forecast scheme for water quality based on the following qualitative classes very good (0–25), good (26–50), poor (51–75), very poor (76–100) and extremely polluted/non-potable (>100). Water quality forecasting by using Artificial Intelligence (AI) is a meaningful method of protecting public health because of its possibility to provide early warning regarding harmful water pollutants.The main objective of the present study is to forecast the WQI time series data based on water physical, chemical and flow status parameters and associated WQ index scores. The Cascade-forward network (CFN) models, along with the Radial Basis Function Network (RBF) as a benchmark model, were developed using data from 2011 to 2017 and WQI forecasts were produced for the period 2018–2019 at all sites. The nineteen input water quality features represent the initial dataset. Moreover, the Random Forest (RF) algorithm refines the initial dataset by selecting eight features considered the most relevant. Both datasets are employed for constructing the predictive models. According to the results of appraisal, the CFN models produced better outcomes (MSE = 0.083/0,319 and R-value 0.940/0.911 in quarter I/quarter IV) than the RBF models. In addition, results show that both the CFN and RBF models could be effective for predicting time series data for water quality when the eight most relevant features are used as input variables. Also, the CFNs provide the most accurate short-term forecasting curves which reproduce the WQI for the first and fourth quarters (the cold season). The second and third quarters presented a slightly lower accuracy. The reported results clearly demonstrate that CFNs successfully forecast the short-term WQI as they may learn historic patterns and determine the nonlinear relationships between the input and output variables.

On the trade-off between profitability, complexity and security of forecasting-based optimization in residential energy management systems

With the emergence of affordable access to data sources, machine learning models and computational resources, sophisticated control concepts for residential energy management systems (EMSs) are on the rise. At the heart of those are production and consumption forecasts. Given the wide spectrum of implementation opportunities, selection of appropriate forecasting strategies is challenging. This work systematically evaluates forecasting-based optimization for residential EMSs in terms of trade-offs between economic profitability, computational complexity and security. The foundation of the study is two real prosumer cases equipped with a photovoltaic-battery system. Results demonstrate that, within the considered scenarios, best trade-offs are achieved based on forecasts of a default gradient-boosted decision trees model, using a short initial training set, weather forecast inputs and regular retraining. Over 90% of the theoretical maximum economic benefit is achieved in this scenario, at significantly lower computational complexity than others with similar savings, while being applicable to new systems without large data history. In terms of security, this scenario exhibits tolerance against weather input manipulation. However, sensitivity to price tampering may require data integrity checking in residential EMSs.

A Heuristic Approach to Forecasting and Selection of a Portfolio with Extra High Dimensions

The performance of a financial portfolio depends on the output of two tasks: first, a forecasting process, where quantities of interest for the investors, such as the rate of return and risk for each stock, are predicted into the future, and second, an optimization process, where those individual stocks are formed into the portfolio optimizing the combined risk and reward features. However, in very large dimensions, when the number of stocks is high, those two quantitative problems often become intractable because of a loss in precision. This paper introduces a forecasting and portfolio formation strategy in multiple periods based on the splitting of the multivariate forecasting model into multiple bivariate forecasting models and updating investment weights at each period based on the predicted target quantities for the returns and the covariances. The methodology proposed is suitable for a very large portfolio of assets. The experimental results are based on a sample of one thousand stocks from the Chinese stock market. For such a large sample, the forecast and optimization process is executed speedily. The investment strategies are benchmarked with the equally weighted portfolio. In the long run, they offer a better investment performance in terms of a higher rate of return or lower risk, compared with this portfolio, demonstrating the applicability and economic value of the proposed methodology in practice.

Cloud-resolving time-lagged rainfall ensemble forecasts for typhoons in Taiwan: Examples of Saola (2012), Soulik (2013), and Soudelor (2015)

As high resolution is required for numerical models to adequately simulate convective storms and thus produce quantitative precipitation forecasts (QPFs), a time-lagged ensemble out to 8 days at 6-h intervals using a 2.5-km cloud-resolving model is applied to three rainy typhoons that made landfall in Taiwan in recent years: Saola (2012), Soulik (2013), and Soudelor (2015), following an earlier study.For the three typhoons where the worst-case rainfall scenario turned out to happen in Taiwan, the system was able to predict this particular scenario with high accuracy (with a rainfall pattern similar to the observed) at an earliest lead time of about 162 h, 79 h, and 164 h before landfall, and thus provided key information for early preparation. Within the short range (≤72 h), as the predicted tracks converged toward the best track, high-quality QPFs were consistently generated starting at about 36 h, 55 h, and 80 h prior to landfall, respectively, with derived probabilities in good to excellent agreement with the observations, even at extreme thresholds (≥500 and 700 mm in 24 h). Leading up to the track convergence, the probabilities across various rainfall thresholds increased markedly, so their time evolution also provided useful information for decision makers and hazard preparation.The underlying reason behind our results is the high predictability of topographic rainfall in Taiwan produced by the typhoon circulation, which cannot be properly captured without high model resolution. Thus, without compromising the resolution, we demonstrate the advantages of the time-lagged strategy for ensemble forecasting, assuming QPF as the key target forecast parameter.

A Monte Carlo simulation and sensitivity analysis framework demonstrating the advantages of probabilistic forecasting over deterministic forecasting in terms of flood warning reliability

Despite the significant progress in probabilistic forecasting science in the last two decades, particularly in the quantification of predictive uncertainty (PU), most operational flood early warning systems (FEWSs) continue to be based on deterministic forecasts. Thereupon, additional work is needed to demonstrate the advantages of using PU over deterministic forecasting to enhance the uptake of probabilistic forecasts in flood warning decision-making. In this paper, a Monte-Carlo (MC)-based sensitivity analysis is done to explore how the outcomes of flood peak water level-based deterministic and -probabilistic warning strategies behave when factors controlling the forecast quality are perturbed. The flood warning reliability is evaluated through the probability of detection (POD) and false alarm ratio (FAR) based on two criteria: a flooding threshold-based criterion (FTC), and a new floodplain property-based criterion (FPC) based on inundation level forecasting. The results of this work show that the advantage of a probabilistic strategy over a deterministic one is greater when PU is relatively high. The probabilistic strategy is robust to biases in the mean and variance of forecasts by maintaining POD and FAR stability, while this is not the case for a deterministic warning strategy. Likewise, it was concluded that if inundation level forecasting is undertaken (FPC), improved forecasts would be needed to achieve the same reliability level as for FTC, reflecting the more demanding FPC criterion. Also, the levels of correlation needed to achieve acceptable operational values of POD and FAR are shown. These results provide new insight into the advantages of a probabilistic forecasting strategy under several forecast quality scenarios and guidance for the design of operational FEWSs.

The true value of a forecast: Assessing the impact of accuracy on local energy communities

Energy communities have become a key component of growing smart grids that integrate distributed renewable energy resources, energy storage technologies, and load management techniques. The random nature of the weather causes challenges for the reliability, power quality, and supply–demand balance of such microgrids. Therefore, energy demand forecasts are increasingly crucial for the effective and continuous operation of the power grid. They also aid in achieving the best possible use of resources to push the limits of self-sufficiency. This study examines not only the quality but the so-called value of a forecast from the point of choosing a forecasting approach. Usually, forecasting approaches are ranked using quality metrics, such as the mean absolute percentage error (MAPE) or root mean square error (RMSE). In addition, the value of a forecast is considered in this study by measuring concrete results for a local energy community (LEC), such as the load cover factor, supply cover factor, on-site energy ratio, and cost of electricity. These evaluations are based on a model of an LEC that includes not only the electric components but also a building and a selected heat pump system for space heating and cooling that is fully dynamical. The optimal operation of this exemplary LEC and the integration of demand forecasts for electricity and domestic hot water (DHW) are achieved with model predictive control (MPC). Several relevant studies on management in LEC are available, but almost none of the publications examine demand forecasting strategies and optimal building asset optimisation at the same time. This research provides two major contributions: first, by developing a more comprehensive framework to assess forecasting performance with reference to energy communities; and second, by highlighting the connection between quality and value indicators of forecasts in the context of energy communities. This study’s findings show that depending solely on quality metrics when choosing a forecasting approach is insufficient and gives no clear statement about the true value of a forecast. This paper identifies the impact of more accurate forecasts on energy community performance measures and attempts to provide an outlook on theoretically achievable improvements based on significantly better forecasts. Finally, this work highlights several open research issues and prospects.

Energy generation forecasting: elevating performance with machine and deep learning

Distribution System Operators (DSOs) and Aggregators benefit from novel Energy Generation Forecasting (EGF) approaches. Improved forecasting accuracy may make it easier to deal with energy imbalances between production and consumption. It also aids operations such as Demand Response (DR) management in Smart Grid architecture. This work aims to develop and test a new solution for EGF. It combines various methodologies running EGF tests on historical data from buildings. The experimentation yields different data resolutions (15 min, one hour, one day, etc.) while reporting accuracy errors. The optimal forecasting technique should be relevant to a variety of forecasting applications in a trial-and-error manner, while utilizing different forecasting strategies, ensemble approaches, and algorithms. The final forecasting evaluation incorporates performance metrics such as coefficient of determination ({R^{2}}), Mean Absolute Error (MAE), Mean Squared Error (MSE) and Root Mean Squared Error (RMSE), presenting a comparative analysis of results.

A bi‐level ensemble learning approach to complex time series forecasting: Taking exchange rates as an example

AbstractForecasting complex time series faces a huge challenge due to its high volatility. To improve the accuracy and robustness of prediction, this paper proposes a bi‐level ensemble learning approach by combining decomposition‐ensemble forecasting and resample strategies. The bi‐level ensemble approach consists of four steps: data decomposition via singular spectrum analysis (SSA), resampling by employing a bagging algorithm, individual forecasting utilizing extreme learning machine (ELM), and introducing sorting‐pruning and simple addition ensemble strategies for integrating the inner and outer results, respectively. To verify the effectiveness of the established forecasting approach, three exchange rate time series are selected as samples. The results reveal that the proposed model is significantly better than the other benchmarks at different lead times, which indicates that it can be regarded as an effective and promising tool for complex time series forecasting.

Can Brazilian Central Bank communication help to predict the yield curve?

AbstractThis paper investigates whether Brazilian Central Bank communication helps to forecast the yield curve. Our forecast strategy involves two steps: First, we analyze textual Central Bank documents to extract sentiment variables that describe its communication, and then, we include those sentiment variables as additional factors into the dynamic Nelson–Siegel term structure model. We found that sentiment variables contain predictive information for yield curve forecasting. Specifically, when combined with macroeconomic variables, the sentiment variables improve the accuracy of the forecast for short maturities and forecast horizons. In addition, sentiment variables are useful in forecasting for medium and long forecast horizons for all maturities. Besides finding a new source of information to forecast the yield curve, the results indicate that the information provided by Central Bank affects market participants, proving to be a useful tool for monetary policy.

Developing Feedforward Neural Networks as Benchmark for Load Forecasting: Methodology Presentation and Application to Hospital Heat Load Forecasting

For load forecasting, numerous machine learning (ML) approaches have been published. Besides fully connected feedforward neural networks (FFNNs), also called multilayer perceptron, more advanced ML approaches like deep, recurrent or convolutional neural networks or ensemble methods have been applied. However, evaluating the added benefit by novel approaches is difficult. Statistical or rule-based methods constitute a too low benchmark. FFNNs need extensive tuning due to their manifold design choices. To address this issue, a structured, comprehensible five-step FFNN model creation methodology is presented, which constitutes of initial model creation, internal parameter selection, feature engineering, architecture tuning and final model creation. The methodology is then applied to forecast real world heat load data of a hospital in Germany. The forecast constitutes of 192 values (upcoming 48 h in 15 min resolution) and is composed of a multi-model univariate forecasting strategy, with three test models developed at first. As a result, the test models show great similarities which simplifies creation of the remaining models. A performance increase of up to 18% between initial and final models points out the importance of model tuning. As a conclusion, comprehensible model tuning is vital to use FFNN models as benchmark. The effort needed can be reduced by the experience gained through repeated application of the presented methodology.

Statistical evaluation of earthquake forecast efficiency using earthquake-catalog and fault slip rate in the Sichuan-Yunnan region, China

Epicenter locations are generally adjacent to active faults and past seismicity regions. Past earthquake catalogs and the geometry of the active faults can provide key prior knowledge concerning earthquake forecasts. In this study, we first introduce two straightforward dedicated models, the proximity-to-past-earthquakes (PPE) and proximity-to-mapped-faults (PMF) models, to fit the seismicity in the Sichuan-Yunnan region, China. The hybrid proximity-to-known-sources (PKS) model with the optimized model parameters is then used to estimate the probability of earthquake occurrence. Second, to compare the PKS forecast efficiency to those of different models, retrospective tests are applied to a dataset located in the Sichuan-Yunnan region. The results show that the probability maps derived from PPE, PMF, and PKS have non-uniform Poisson distribution features and that there is forecasting significance for past cases of moderate earthquakes in the test region. Finally, using Molchan error diagram tests, we find that the hybrid PKS model performs better than the other models in the testing region. The unsatisfactory performance of the PMF model for earthquake forecasting may lie both in the incompleteness of the fault database and the lack of consideration of co-seismic ruptures. Therefore, one of the three models can be used as a base model for comparing and evaluating earthquake forecast strategies.

Comparison of five strategies for seasonal prediction of bioclimatic indicators in the olive sector

This paper assesses the forecast quality of five seasonal forecasting strategies applied to different bioclimatic indicators tailored to the olive sector. In total, five indicators have been selected considering their importance in the management of the olive orchard. As time progresses through the indicator target period, the impact of the increasing share of actual observations included in its computation has been evaluated by examining the variabilities of correlation coefficients and fair rank probability skill scores for each initialization date. The results show that blending either seasonal predictions or climatology with observations enhanced the capability of forecasting the tercile category for all the indicators when compared to the use of climatology or seasonal predictions alone. In fact, for Spring Maximum Temperature and Growing Season Temperature indicators, the combination of observations and SEAS5 predictions could outperform the other methods for most of the start months. As for those threshold-defined indicators, namely Spring Heat Days and Summer Heat Stress Days, the end-users are highly encouraged to use climatology in the first month and combine it with observations as soon as the latter becomes available.

An improved Wavenet network for multi-step-ahead wind energy forecasting

Accurate multi-step-ahead wind speed (WS) and wind power (WP) forecasting are critical to the scheduling, planning, and maintenance of wind farms. Previous forecasting methods tend to focus on improving forecast accuracy by integrating different models and disaggregating data while neglecting the forecasting ability of basic models. In addition, traditional multi-step-ahead output strategies have limitations that constrain the forecasting capability of models. To overcome the above challenges, this study proposes a novel forecasting model called ED-Wavenet-TF. It adopts two Wavenet networks as Encoder and Decoder connected by the multi-head self-attention mechanism. And, teacher forcing is used as the multi-step-ahead output strategy for WS and WP forecasting. In the training phase, ED-Wavenet-TF uses a portion of the actual data to correct the errors at the intermediate forecasting steps, while in the forecasting phase, it runs through an inference loop to make forecasts. In this study, two WS datasets and two WP datasets are used to validate the performance of ED-Wavenet-TF with univariate input. The results show that compared with Wavenet, the symmetric mean absolute percentage error of ED-Wavenet-TF at four forecasting steps is lower by at least 4.8577% on average for the WS datasets and 8.9463% on average for the WP datasets. The advantages of ED-Wavenert-TF over ten comparable models are confirmed by four evaluation indicators and the Harvey, Leybourne, and Newbold statistical hypothesis test. Moreover, ED-Wavenet-TF is extended to make multi-step-ahead forecasts with multivariate inputs, whose effectiveness is demonstrated on another open WS dataset.