Forecasters Research Articles

A natural gas consumption forecasting system for continual learning scenarios based on Hoeffding trees with change point detection mechanism

Forecasting natural gas consumption, considering seasonality and trends, is crucial in planning its supply and consumption and optimizing the cost of obtaining it, mainly by industrial entities. However, in times of threats to its supply, it is also a critical element that guarantees the supply of this raw material to meet individual consumers’ needs, ensuring society’s energy security. This article introduces a novel multistep forecasting of natural gas consumption with change point detection integration for model collection selection with continual learning capabilities using data stream processing. The performance of the forecasting models based on the proposed approach is evaluated in a complex real-world use case of natural gas consumption forecasting. Furthermore, the methodology generability was verified in an electricity load forecasting task. We employed Hoeffding tree predictors as forecasting models and the Pruned Exact Linear Time (PELT) algorithm for the change point detection procedure. The change point detection integration enables the selection of a different model collection for successive time frames. Thus, three model collection selection procedures are defined and evaluated for forecasting scenarios with various densities of detected change points. These models were compared with change point agnostic baseline approaches and deep learning models. Our experiments show that the proposed approach provides superior results to deep learning models for both datasets and that fewer change points result in a lower forecasting error regardless of the model collection selection procedure employed.

Calibration of Typhoon Track Forecasts Based on Deep Learning Methods

An accurate forecast of typhoon tracks is crucial for disaster warning and mitigation. However, existing numerical weather prediction models, such as the Weather Research and Forecasting (WRF) model, still exhibit significant errors in track forecasts. This study aims to improve forecast accuracy by correcting WRF-forecasted tracks using deep learning models, including Bidirectional Long Short-Term Memory (BiLSTM) + Convolutional Long Short-Term Memory (ConvLSTM) + Wide and Deep Learning (WDL), BiLSTM + Convolutional Gated Recurrent Unit (ConvGRU) + WDL, and BiLSTM + ConvLSTM + Extreme Deep Factorization Machine (xDeepFM), with a comparison to the Kalman Filter. The results demonstrate that the BiLSTM + ConvLSTM + WDL model reduces the 72 h track prediction error (TPE) from 255.18 km to 159.23 km, representing a 37.6% improvement over the original WRF model, and exhibits significant advantages across all evaluation metrics, particularly in key indicators such as Bias2, Mean Squared Error (MSE), and Sequence. The decomposition of MSE further validates the importance of the BiLSTM, ConvLSTM, WDL, and Temporal Normalization (TN) layers in enhancing the model’s spatio-temporal feature-capturing ability.

Multistage stochastic programming for integrated network optimization in hurricane relief logistics and evacuation planning

AbstractIn this article, we study the integrated hurricane relief logistics and evacuation planning (IHRLEP) problem, integrating hurricane evacuation and relief item pre‐positioning operations that are typically treated separately. We propose a fully adaptive multistage stochastic programming (MSSP) model and solution approaches based on two‐stage stochastic programming (2SSP). Utilizing historical forecast errors modeled using the auto‐regressive model of order one, we generate hurricane scenarios and approximate the hurricane process as a Markov chain, and each Markovian state is characterized by the hurricane's location and intensity attributes. We conduct a comprehensive numerical experiment based on case studies motivated by Hurricane Florence and Hurricane Ian. Through the computational results, we demonstrate the value of fully adaptive policies given by the MSSP model over static ones given by the 2SSP model in terms of the out‐of‐sample performance. By conducting an extensive sensitivity analysis, we offer insights into how the value of fully adaptive policies varies in comparison to static ones with key problem parameters.

A predictive fuzzy logic and rule-based control approach for practical real-time operation of urban stormwater storage system

Predictive real-time control (RTC) strategies are usually more effective than reactive strategies for the intelligent management of urban stormwater storage systems. However, it remains a challenge to ensure the practicality of RTC strategies that use accessible, non-idealized predictive information while improving their efficiency for successive rainfall events instead of specific phases. This study developed a predictive fuzzy logic and rule-based control (PFL-RBC) approach to address the continuous control of individual storage systems. This approach incorporates total rainfall depth forecast information with an intra-storm fuzzy logic system to optimize peak flow control and several rule-based strategies for pre-storm water detention, reuse, and release control. Computational experiments were conducted using a storage tank case study to test the proposed approach under various rainfall conditions and storage sizes. The results showed that PFL-RBC outperformed static rule-based control in infrequent design storms and realistic continuous rainfall events, reducing flood peaks and volumes by 55 %∼87 % and 7 %∼20 %, respectively, and significantly increasing water detention time and reuse volume. Meanwhile, PFL-RBC required less predictive information to achieve a 6 %∼15 % advantage in peak flow control compared to optimized model predictive control. More importantly, PFL-RBC was reliable in the face of input uncertainty, with <25 % performance loss for water quantity control when the realistic forecast error ranged from -50 % to +50 %. These findings suggest that the proposed approach has great potential to enhance the efficiency and practicality of stormwater storage operations.

Public debt determinants: A time-varying analysis of core and peripheral Euro area countries

This study employs a Bayesian Interacted Panel VAR model to estimate time-varying Generalized Forecast Error Variance Decomposition, analyzing how key determinants affect debt in Core and Peripheral Euro Area countries. Results highlight varying effects of determinants across periods and subgroups.

Time‐series‐based ensemble model output statistics for temperature forecasts postprocessing

AbstractThe uncertainty in numerical weather prediction models is nowadays quantified by the use of ensemble forecasts. Although these forecasts are continuously improved, they still suffer from systematic bias and dispersion errors. Statistical postprocessing methods, such as the ensemble model output statistics (EMOS), have been shown to substantially correct the forecasts. This work proposes an extension of EMOS in a time‐series framework. Besides taking account of seasonality and trend in the location and scale parameter of the predictive distribution, the autoregressive process in the mean forecast errors or the standardized forecast errors is considered. The models can be further extended by allowing generalized autoregressive conditional heteroscedasticity. Furthermore, it is outlined how to use these models for arbitrary forecast horizons. To illustrate the performance of the suggested EMOS models in time‐series fashion, we present a case study for the postprocessing of 2 m surface temperature forecasts using five different lead times and a set of observation stations in Germany. The results indicate that the time‐series EMOS extensions are able to significantly outperform the benchmark models EMOS and autoregressive EMOS (AR‐EMOS) in most of the lead time–station cases.

Research on the risk spillover effect of carbon emission trading market and various industry markets in China and the European Union

AbstractWidely concerns over the carbon emission problems have been aroused. Prior studies have documented the correlation between the carbon emission market and industries separately. This study compared the tail risks and risk spillover effect of the carbon emission and 11 industries markets in China and the European Union (EU) by using the multivariate multi‐quartile conditional autoregressive at‐risk model. Moreover, to evaluate the risk spillover of each market under extreme risk conditions in time domain and frequency domain, DY spillover index and BK spillover index were constructed via generalized forecast error variance decomposition and generalized causation spectrum, respectively. Findings are as follows: (1) The tail risks and fluctuation of the trend of Chinese industry markets reflects more higher and larger than those in the EU; (2) The EU suffers from smaller external shocks, while China has the opposite result and can recover relatively faster; (3) In China, energy, industrial, information technology, financial, real estate, consumer goods, carbon emissions, and discretionary consumption industries are risk spillover industries, while healthcare, materials, telecommunication services, and utilities industries are risk receiving industries. In contrast, the risk spillover industries in the EU remain consistent with those of the Chinese markets except for the materials, discretionary consumption, consumer goods, information technology and real estate industries. (4) On the short‐term, medium‐term and long‐term scales, the risk spillover of China's carbon emission trading market and various industries is basically consistent with that of the EU. These findings contribute to reducing greenhouse gas emissions and achieving the goal of carbon peak and carbon neutrality.

Study on Downscaling Correction of Near-Surface Wind Speed Grid Forecasts in Complex Terrain

Accurate forecasting of wind speeds is a crucial aspect of providing fine-scale professional meteorological services (such as wind energy generation and transportation operations etc.). This article utilizes CMA-MESO model forecast data and CARAS-SUR_1 km ground truth grid data from January, April, July, and October 2022, employing the random forest algorithm to establish and evaluate a downscaling correction model for near-surface 1 km resolution wind-speed grid forecast in the complex terrain area of northwestern Hebei Province. The results indicate that after downscaling correction, the spatial distribution of grid forecast wind speeds in the entire complex terrain study area becomes more refined, with spatial resolution improving from 3 km to 1 km, reflecting fine-scale terrain effects. The accuracy of the corrected wind speed forecast significantly improves compared to the original model, with forecast errors showing stability in both time and space. The mean bias decreases from 2.25 m/s to 0.02 m/s, and the root mean square error (RMSE) decreases from 3.26 m/s to 0.52 m/s. Forecast errors caused by complex terrain, forecast lead time, and seasonal factors are significantly reduced. In terms of wind speed categories, the correction significantly improves forecasts for wind speeds below 8 m/s, with RMSE decreasing from 2.02 m/s to 0.59 m/s. For wind speeds above 8 m/s, there is also a good correction effect, with RMSE decreasing from 2.20 m/s to 1.65 m/s. Selecting the analysis of the Zhangjiakou strong wind process on 26 April 2022, it was found that the downscaled corrected forecast wind speed is very close to the observed wind speed at the station and the ground truth grid points. The correction effect is particularly significant in areas affected by strong winds, such as the Bashang Plateau and valleys, which has significant reference value.

Credit Cycles, Expectations, and Corporate Investment

Abstract We provide a systematic empirical assessment of the Minsky hypothesis that business fluctuations stem from irrational swings in expectations. Using predictable firm-level forecast errors, we build an aggregate index of irrational expectations and use it to provide three sets of results. First, we show that our index predicts aggregate credit cycles. Next, we show that these predictable credit cycles drive cycles in firm-level debt issuance and investment and similar cycles between financially constrained and unconstrained firms, as Minsky predicts. Finally, we show more pronounced cycles in firm-level financing and investment for firms with ex ante more optimistic expectations. (JEL G31, G32, G40, E32, E44)

Evaluating the influence of Nano-GO concrete pavement mechanical properties on road performance and traffic safety using ANN-GA and PSO techniques

The burgeoning demand for durable and eco-friendly road infrastructure necessitates the exploration of innovative materials and methodologies. This study investigates the potential of Graphene Oxide (GO), a nano-material known for its exceptional dispersibility and mechanical reinforcement capabilities, to enhance the sustainability and durability of concrete pavements. Leveraging the synergy between advanced artificial intelligence techniques—Artificial Neural Networks (ANN), Genetic Algorithms (GA), and Particle Swarm Optimization (PSO)—it is aimed to delve into the intricate effects of Nano-GO on concrete's mechanical properties. The empirical analysis, underpinned by a comparative evaluation of ANN-GA and ANN-PSO models, reveals that the ANN-GA model excels with a minimal forecast error of 2.73%, underscoring its efficacy in capturing the nuanced interactions between GO and cementitious materials. An optimal concentration is identified through meticulous experimentation across varied Nano-GO dosages that amplify concrete's compressive, flexural, and tensile strengths without compromising workability. This optimal dosage enhances the initial strength significantly, and positions GO as a cornerstone for next-generation premium-grade pavement concretes. The findings advocate for the further exploration and eventual integration of GO in road construction projects, aiming to bolster ecological sustainability and propel the adoption of a circular economy in infrastructure development.

Traffic Prediction as a Multidimensional Random Process in a Three-Dimensional High-Density Internet of Things Network

Relevance. The relevance of the topic considered in the article lies in the active transformation of communication networks and the formation of a three-dimensional high-density communication network, which changes the structure of data traffic, therefore, for this type of network, a traffic model is considered as a multidimensional random process. The main purpose of the study is to improve the efficiency of network traffic forecasting by developing a method, the distinctive feature of which is traffic forecasting as a multidimensional random process, taking into account the mutual dependence of individual flows produced by network nodes. Methods. The paper considers an algorithm for training an artificial neural network (ANN) based on the method of reducing the root of the mean square error RMSE, and also proposes forecasting methods using LSTM-type ANNs and adapting model parameters to changing network operating conditions. The use of LSTM-type ANN for forecasting a multivariate random process describing traffic in a three-dimensional high-density network can yield better results than forecasting individual traffic flows as independent random processes due to the consideration of mutual influences between different traffic flows.The results. Building the corresponding model, collecting statistics (obtaining a training sample), training the ANN and performing the forecast require the use of computing resources. Thus, the forecasting efficiency can be defined as a decrease in the forecasting error while maintaining the volume of resources used or a decrease in the volume of resources while maintaining the forecasting error. In the course of solving the scientific problem, criteria were identified for selecting the value of a unit interval (lag), which, together with the forecasting interval, significantly affects the final scenario.The theoretical significance The scientific novelty of the work lies in the assessment of the change in the error in forecasting the traffic of a three-dimensional high-density communication network as a multivariate random process, compared to presenting the forecast of the same traffic as a set of independent random processes.Significance (theoretical). The efficiency of traffic forecasting as a multidimensional random process in a three-dimensional high-density communication network increases with increasing dimensionality. Thus, such traffic in forecasting problems should be considered as a multidimensional random process, the dimensionality of which is equal to the number of network nodes producing traffic.Significance (practical). The results obtained in the work can be used in the future to optimize the functioning of the traffic management system.

A reliable ensemble forecasting modeling approach for complex time series with distributionally robust optimization

Accurate ensemble forecast results provide strong support for management decisions. While the existing ensemble forecasting model ignores the requirement of directional accuracy making its prediction direction is usually error-prone. To address this issue, we develop a convex directional prediction error measure and subsequently construct a reliable ensemble forecasting model that minimizes the horizontal prediction error with the bounded directional prediction error. Mathematically, we provided the proper range of the required directional error level. Furthermore, we find the classical ensemble forecasting model is a special case of our study when the directional error level is larger than the upper bound we gave in this study. To deal with the possible deterioration of the individual model over time, we also considered its worst possible prediction performance by introducing the distributionally robust optimization (DRO) technique into the proposed reliable ensemble forecasting model. Technically, we showed that the DRO-based reliable ensemble forecasting model is convex and can be reformulated into a second-order cone problem which can be readily solved by off-the-shelf solvers. Finally, the effectiveness of the proposed reliable ensemble forecasting model and the DRO-based reliable ensemble forecasting model were validated on three different datasets, e.g., the exchange rate, the oil price, and the country risk index. To sum up, we construct a reliable ensemble forecasting model to simultaneously control the horizontal prediction error and directional prediction error of ensemble forecasting, and thus enhance the reliability of ensemble forecasting.

Revisiting the moral forecasting error – A preregistered replication and extension of “Are we more moral than we think?”

Predictions are often inaccurate. Still, the direction of prediction errors may vary. Contrary to research on the intention-behavior gap, where people fail to live up to their ambitions, a study on “moral forecasting” found that people behaved more honestly than they predicted. In this registered report, we present two close replication attempts and one conceptual replication attempt of this moral forecasting error across two experiments. In Experiment 1 (N = 1839), we recruited a general population sample from the same country as the original study (Canada) to an online experiment. We successfully replicated the moral forecasting error using a math-based cheating task from the original study: Predicted cheating was much higher in a moral forecasting condition than actual cheating in a moral action condition (d = 0.69). In Experiment 2 (N = 1381) we replicated the forecasting error again, using the same task in a general population sample from the U.S. (d = 0.72). However, we were unable to conceptually replicate the effect using a different dishonesty measure, the “mind game”, in Experiment 1 (φ = 0.03). We also could not reduce the forecasting error through a debiasing intervention in Experiment 2 (d = 0.01). Across both experiments, participants predicted that others would cheat much more than they would themselves. In this registered report, we conclude that the moral forecasting error is robust for the original cheating task. We also show that it can generalize contextually (from a lab to an online setting), but not to a different task. Future research may show exactly when predictions about one's own honesty are pessimistic rather than optimistic.

A Physical‐Informed Neural Network for Improving Air‐Sea Turbulent Heat Flux Parameterization

AbstractThe parameterizations of air‐sea turbulent heat flux are one of the major bottlenecks in atmosphere‐ocean coupled model development, which play a crucial role in sea surface temperature (SST) prediction. Recently, neural networks start to be applied for the development of parameterizations of interface turbulent heat flux. However, these new parameterizations are primairily developed for specific regions and have not been tested in real atmosphere‐ocean coupled models. In this study, we propose a new air‐sea heat flux parameterization using a physical‐informed neural network (PINN) based on multiple observational data sets worldwide. Evaluated with an independent observation data set, it is shown that the PINN can significantly reduce the RMSE of latent heat flux by at least about 48.6% compared to three traditional bulk formulas. Moreover, the PINN can be flexibly updated with new observational data by transfer learning. To test the performance of the new parameterization in realistic application, we implement the PINN into a global ocean‐atmosphere coupled model and make seasonal forecasts for the first time. The PINN markedly reduce the errors of equatorial SST forecast, indicating a good performance of the PINN‐based air‐sea turbulent heat flux scheme. Noticeably, due to limited observational data, the NN‐based parameterizations tend to underestimate heat flux at high wind speeds compared with bulk formula‐based parameterizations. With more data available at extreme conditions, the PINN can be improved via transfer learning and need to be futher evaluated. This study suggests that PINN‐based air‐sea heat flux parameterization is promising to improve SST simulation.

An extension to ensemble forecast of conditional nonlinear optimal perturbation considering nonlinear interaction between initial and model parametric uncertainties

Initial and model uncertainties are the main sources of forecast errors, making the single and deterministic forecasts unreliable. To estimate these uncertainties, a growing consensus shifts towards ensemble forecasting, aiming to provide the probability density distribution of the atmosphere. However, current ensemble methods either focus on single-source uncertainties or employ a simple superposition of the two, neglecting the nonlinear interaction between them, and thus fail to reflect the real forecast uncertainty. Motivated by this, this study extends the CNOP approach, defined as the optimal growth considering nonlinear interaction between initial and model parameters, to the scenario of ensemble forecasts and proposes an orthogonal CNOPs method (O-CNOP-IPs). This method concerns the nonlinear effect of initial and model parametric uncertainties through a joint optimization strategy and enhances the estimation of this effect by providing diversity and independent CNOPs (via orthogonality). To evaluate the performance of O-CNOP-IPs, extensive experiments are conducted for North Atlantic Oscillation (NAO) ensemble forecasts in the realistically configured Community Earth System Model (CESM). Our findings reveal that the O-CNOP-IPs method outperforms existing methods in forecast skill and reliability, improving deterministic skill by 17.5 % and probabilistic skill by 52 %–63 %. Our dynamic analysis also unveils that this method undergoes rapid development in the early stage and effectively neutralizes errors in control forecasts, significantly enhancing the reliability of ensemble forecasts. It is expected that O-CNOP-IPs plays a significant role in accurately representing the forecast uncertainty of other high-impact weather and climate phenomena.

Evaluation of Subseasonal Forecast Skill for Northern Hemisphere Winter Snow Cover

Abstract Accurate subseasonal forecasts for snow cover have significant socioeconomic value. This paper evaluates subseasonal forecasts for winter snow cover in the Northern Hemisphere as predicted by three numerical models: the Model for Prediction Across Scales–Atmosphere (MPAS-A), the China Meteorological Administration (CMA) model, and the European Centre for Medium-Range Weather Forecasts (ECMWF) model. While these models can generally simulate the spatial distribution of winter snow cover climatology and subseasonal variability, they tend to underestimate both the climatology and the intensity of subseasonal variability. Compared to persistence forecasts, these models demonstrate skill in subseasonal snow cover forecasting. Notably, the ECMWF model outperforms the MPAS and CMA models. The sensitivity of the surface air temperature subseasonal forecast skill to the predicted snow cover was also investigated using the MPAS. The results show that for forecasts with lead times of 1–2 weeks, the predicted snow cover contributes to the temperature forecasting skill. However, for forecasts with lead times of 3–4 weeks, the predicted snow cover does not enhance the temperature forecasting skill. Furthermore, part of the errors in temperature forecasts can be attributed to inaccuracies in snow cover forecasts with lead times of 2 weeks or more. These findings suggest that refining snow cover parameterization schemes and effectively exploiting predictability from snow cover can enhance the skill of subseasonal atmospheric forecasts. Significance Statement Snow cover is a crucial variable in hydrometeorology. Subseasonal forecasting, which involves predicting snow cover anomalies 1–4 weeks in advance, has socioeconomic value. We conducted an evaluation of the subseasonal forecasts for Northern Hemisphere winter snow cover produced by three numerical models. This evaluation provides insights into the accuracy and reliability of these models, which could contribute to their enhancement. Furthermore, we examined the impact of the predicted snow cover on the skill of surface air temperature subseasonal forecasts. The results suggest that improvements in snow cover modeling and forecasting can lead to more accurate subseasonal atmospheric forecasts. Therefore, future efforts to refine snow cover parameterization schemes suitable for subseasonal forecasting are promising and worthwhile.

Leveraging Deterministic Weather Forecasts for In Situ Probabilistic Temperature Predictions via Deep Learning

Abstract We propose a neural network approach to produce probabilistic weather forecasts from a deterministic numerical weather prediction. Our approach is applied to operational surface temperature outputs from the Global Deterministic Prediction System up to 10-day lead times, targeting METAR observations in Canada and the United States. We show how postprocessing performance is improved by training a single model for multiple lead times. Multiple strategies to condition the network for the lead time are studied, including a supplementary predictor and an embedding. The proposed model is evaluated for accuracy, spread, distribution calibration, and its behavior under extremes. The neural network approach decreases the continuous ranked probability score (CRPS) by 15% and has improved distribution calibration compared to a naive probabilistic model based on past forecast errors. Our approach increases the value of a deterministic forecast by adding information about the uncertainty, without incurring the cost of simulating multiple trajectories. It applies to any gridded forecast including the recent machine learning–based weather prediction models. It requires no information regarding forecast spread and can be trained to generate probabilistic predictions from any deterministic forecast. Significance Statement Weather is difficult to predict a long time in advance because we cannot measure the state of the atmosphere precisely enough. Consequently, it is common practice to run forecasts several times and look at the differences to evaluate how uncertain the prediction is. This process of running ensemble forecasts is expensive and consequently not always feasible. We propose a middle ground where we add uncertainty information to forecasts that were run only once, using artificial intelligence. Our method increases the value of these forecasts by adding information about the uncertainty without incurring the cost of multiple full simulations.

Adaptive formulation for probabilistic storm surge predictions through sharing of numerical simulation results across storm advisories

As a tropical storm/cyclone approaches landfall, real-time probabilistic predictions for the anticipated surge provide valuable information for emergency preparedness/response decisions. These probabilistic predictions are made through an uncertainty quantification process that involves: (i) generating a sufficiently large ensemble of storm scenarios based on the nominal storm advisory and the anticipated forecast errors; (ii) performing high-fidelity numerical simulations to obtain surge predictions for each storm scenario; and (iii) estimating surge statistics of interest by assembling the simulation results. This process is repeated whenever the nominal storm advisory is updated. The number of storm scenarios utilized in the analysis directly impacts the statistical accuracy of the probabilistic predictions; a larger ensemble improves accuracy but requires greater computational resources to provide predictions with the desired expediency to guide real-time decisions. This paper revisits two recently proposed Monte-Carlo (MC) frameworks that aim to improve accuracy without increasing the computational burden: adaptive importance sampling (AIS) and adaptive multi-fidelity Monte Carlo (AMFMC). The foundational concept behind them is similar: share numerical simulation results across the probabilistic predictions performed for different storm advisories to accelerate the MC estimation. This is achieved differently for each approach, through adaptive development of an importance sampling proposal density (for AIS) or a surrogate model (for AMFMC). Here, a direct comparison between these frameworks is established, focusing on the mechanisms for the information sharing and the challenges encountered in tuning the algorithm adaptive characteristics to provide probabilistic estimates across a large number of quantities of interest (QoIs), corresponding to the surge predictions for different locations within the coastal region of interest. As this large number results in conflicting choices for the adaptive characteristics, a compromise solution needs to be promoted. The efficacy of the two frameworks is examined in detail in this setting, comparing the accuracy of idealized implementations (adaptive decisions independently made for each QoI) to the accuracy of practical implementations (single, compromise decision within the MC implementation). The study also showcases the importance of information sharing across storm advisories in real-time probabilistic storm surge predictions and provides guidelines for an efficient adaptive MC formulation in such settings.

Analysis of model error in forecast errors of extended atmospheric Lorenz 05 systems and the ECMWF system

Analysis of model error in forecast errors of extended atmospheric Lorenz 05 systems and the ECMWF system

Peramalan Cadangan Devisa Menggunakan Metode Double Smoothing Exponential dan Metode Fuzzy Time Series

This study aims to forecast the amount of foreign exchange reserves using two methods, namely Double Smoothing Exponential and Fuzzy Time Series. This foreign exchange reserve data is time series data, namely monthly data starting from January 2017 to February 2024 with a total of 86 monthly data. Furthermore, to see the level of accuracy of the forecast results, the Mean Absolute Percentage Error (MAPE) and MSE values ​​are used where the smaller the error value produced, the more accurate the forecast results from the Double Smoothing Exponential Holt and Fuzzy Time Series Cheng methods. The results of forecasting the amount of foreign exchange reserves in March 2024 using the Holt method are 144434.19 and for the Cheng method are 140008.032. The double exponential smoothing method has an MSE value of 8625715.609 and a MAPE value of 1.69%. Meanwhile, with the fuzzy time series method, the MSE value is 11651319.400 and the MAPE value is 1.96%. Forecasting using these two methods has a very good level of accuracy, because it has a MAPE value below 10%. Based on the results of this analysis, it can be seen that the double exponential smoothing method from Holt has a smaller forecast error size than the Cheng fuzzy time series method. Thus, it can be concluded that the Double Exponential Smoothing method is a better method for predicting the amount of foreign exchange reserves.