Rolling Forecast Research Articles

Dynamical system prediction from sparse observations using deep neural networks with Voronoi tessellation and physics constraint

Despite the success of various methods in addressing the issue of spatial reconstruction of dynamical systems with sparse observations, spatio-temporal prediction for sparse fields remains a challenge. Existing Kriging-based frameworks for spatio-temporal sparse field prediction fail to meet the accuracy and inference time required for nonlinear dynamic prediction problems. In this paper, we introduce the Dynamical System Prediction from Sparse Observations using Voronoi Tessellation (DSOVT) framework, an innovative methodology based on Voronoi tessellation which combines convolutional encoder–decoder (CED) and long short-term memory (LSTM) and utilizing Convolutional Long Short-Term Memory (ConvLSTM). By integrating Voronoi tessellations with spatio-temporal deep learning models, DSOVT is adept at predicting dynamical systems with unstructured, sparse, and time-varying observations. CED-LSTM maps Voronoi tessellations into a low-dimensional representation for time series prediction, while ConvLSTM directly uses these tessellations in an end-to-end predictive model. Furthermore, we incorporate physics constraints during the training process for dynamical systems with explicit formulas. Compared to purely data-driven models, our physics-based approach enables the model to learn physical laws within explicitly formulated dynamics, thereby enhancing the robustness and accuracy of rolling forecasts. Numerical experiments on real sea surface data and shallow water systems clearly demonstrate our framework’s accuracy and computational efficiency with sparse and time-varying observations.

Long-term rolling prediction of transformer power load capacity based on the informer model

Abstract There are currently no methods capable of long-term forecasting based on very long-term real-world data. Any false prediction may damage the electrical transformer. For this problem, a transformer power load based on the Informer model is used as a method of long-term rolling forecasting. This method uses the self-attention distillation mechanism in the Informer model to allow the decoder of each layer to shorten the length of the input sequence by half, thus greatly saving the encoder memory overhead, and this paper adds a rolling long-term prediction function, making the prediction decoding time extremely short. Taking the power load data of a certain province in China as a test case, the improved model Informer* was compared with the traditional Informer model. The results show that the prediction accuracy of the Informer* model is higher and the load prediction accuracy is effectively improved.

Rolling forecast of snowmelt floods in data-scarce mountainous regions using weather forecast products to drive distributed energy balance hydrological model

Snowmelt flood forecasting plays a crucial role in water resource management and utilization in arid regions. However, the accuracy of current snowmelt flood forecasts falls short of requirements due to complex mountainous terrain and limited observational data. In this study, We conducted a snowmelt flood simulation and forecasting experiment in a mountainous watershed in northern Xinjiang. First, we used site observations, ERA5-land reanalysis data and WorldClim data to produce historical meteorological forcing, calibrated the VIC-CAS model and improved routing model using historical observed runoff data, and conducted a detailed simulation of the hourly snowmelt flood process in the basin. Then, the latest grid forecast weather product developed by the China Meteorological Administration was integrated into the system to conduct a rolling forecast test of the snowmelt floods in the next 72 h during the snowmelt period in 2023. The remotely sensed snow cover and the freezing level measured by the sounding balloon were assimilated into the model in real time to correct the deviation of runoff simulation. The results show that during the model calibration period, the KGE of the simulated monthly runoff in the Kayertes River basin (KRB) from 2006 to 2011 was 0.83 and the NSE was 0.86. In the continuous hourly runoff simulation during the 2022 snowmelt period, the model well captured the intraday fluctuations of snowmelt runoff in the KRB, with the daily peak flow error within ± 20 % and the peak occurrence time controlled within ± 4 h. In the rolling forecast test in 2023, the daily peak flow error is within ± 20 % (the peak occurrence error during the peak period is even controlled within ± 5 %), and the peak occurrence time is within ± 4 h. Real-time assimilation of freezing level measured by sounding balloons and remotely sensed snow cover can effectively correct the deviations in simulated temperature and snow cover. This study can provide insights for runoff prediction in other high-altitude mountainous areas where data are scarce.

Predicting the Intensity of Tropical Cyclones over the Western North Pacific Using a Dual-Branch Spatiotemporal Attention Convolutional Network

Abstract This paper proposes a spatiotemporal attention convolutional network (STAC-Pred) that leverages deep learning techniques to model the spatiotemporal features of tropical cyclones (TCs) and enable real-time prediction of their intensity. The proposed model employs dual branches to concurrently extract and integrate features from intensity heatmaps and satellite cloud imagery. Additionally, a residual attention (RA) module is integrated into the three-channel cloud imagery convolution process to automatically respond to high wind speed regions. TC’s longitude, latitude, and radius of winds are injected into the multi-timepoint prediction model to assist in the prediction task. Furthermore, a rolling mechanism (RM) is employed to smooth the fluctuation of losses, achieving accurate prediction of TC intensity. We use several TC records to evaluate and validate the universality and effectiveness of the model. The results indicate that STAC-Pred achieves satisfactory performance. Specifically, the STAC-Pred model improves prediction performance by 47.69% and 28.26% compared to the baseline (official institutions) at 3- and 6-h intervals, respectively. Significance Statement Tropical cyclones are one of the most deadly and damaging natural disasters in coastal areas worldwide. Early prediction can significantly reduce casualties and property losses. This study innovatively conducts dimensionality augmentation on one-dimensional intensity numerical sequences and proposes a new network model for rolling forecast of their future intensity. The proposed prototype model (not yet incorporating any atmospheric conditions) shows promising results for 3- and 6-h advance forecasts, providing valuable guidance for forecasters regarding real-time operational predictions of short-term tropical cyclone intensity.

Deep learning-based forecasting model for chlorophyll-a response to tropical cyclones in the Western North Pacific

Tropical cyclones cause increases in sea surface chlorophyll-a concentration, which is important for studying variations in the regional marine environment. Precisely forecasting the variations of sea surface chlorophyll-a concentration induced by tropical cyclones remains a challenge. In this research, a bidirectional long short-term memory (BiLSTM) neural network deep learning model was applied to predict the variations of sea surface chlorophyll-a concentration induced by typhoons in the Western North Pacific (WNP). Typhoons occurring between 2011 and 2020 were used as training cases and those from 2021 to 2022 as forecasting and testing cases. The input variables of the deep learning model include the sea surface wind at 10 meters (U10 and V10), sea surface temperature anomaly (SSTA), and sea surface chlorophyll-a concentration. The output variable was the chlorophyll-a concentration one day after the passage of the typhoon. Data from the previous 7 days were used to predict the chlorophyll-a concentration one day after the typhoon's passage, and the rolling forecast method was employed to predict chlorophyll-a concentration in the following 7 days. To assess the impact of input variables on the model's forecasting performance, ablation experiments were conducted. The results showed that when using U10, V10, and chlorophyll-a from the previous seven days as inputs, the model exhibited the best overall forecasting performance. Taking Typhoons Chanthu, In-fa, and Malou as examples, the root mean square error (RMSE) for the forecast results are 0.0143 mg · m−3, 0.0087 mg · m−3, and 0.0030 mg · m−3, the mean absolute errors (MAE) are 0.0072 mg · m−3, 0.0074 mg · m−3, and 0.0025 mg · m−3, and the spatial anomaly correlation coefficients (ACC) are 0.9968, 0.9775, and 0.9721, respectively. The results reveal that the most accurate forecasting performance was observed during the mid-phase of the moderate-intensity Typhoon Muifa, with RMSE, MAE, and ACC values of 0.0040 mg · m−3, 0.0032 mg · m−3, and 0.9894, respectively. The BiLSTM neural network model had the best forecasting performance for typhoons of moderate intensity and during the mid-term phase. This is because moderate-intensity typhoons or the mature phase of any typhoon tend to have relatively stable and more predictable paths, resulting in better predictions of chlorophyll-a concentrations. In future work, we intend to extend our training and forecasting to typhoons of various intensities, aiming to further refine and enhance predictive performance.

Development of a novel dynamic nosocomial infection risk management method for COVID-19 in outpatient settings

BackgroundApplication of accumulated experience and management measures in the prevention and control of coronavirus disease 2019 (COVID-19) has generally depended on the subjective judgment of epidemic intensity, with the quality of prevention and control management being uneven. The present study was designed to develop a novel risk management system for COVID-19 infection in outpatients, with the ability to provide accurate and hierarchical control based on estimated risk of infection.MethodsInfection risk was estimated using an auto regressive integrated moving average model (ARIMA). Weekly surveillance data on influenza-like-illness (ILI) among outpatients at Xuanwu Hospital Capital Medical University and Baidu search data downloaded from the Baidu Index in 2021 and 22 were used to fit the ARIMA model. The ability of this model to estimate infection risk was evaluated by determining the mean absolute percentage error (MAPE), with a Delphi process used to build consensus on hierarchical infection control measures. COVID-19 control measures were selected by reviewing published regulations, papers and guidelines. Recommendations for surface sterilization and personal protection were determined for low and high risk periods, with these recommendations implemented based on predicted results.ResultsThe ARIMA model produced exact estimates for both the ILI and search engine data. The MAPEs of 20-week rolling forecasts for these datasets were 13.65% and 8.04%, respectively. Based on these two risk levels, the hierarchical infection prevention methods provided guidelines for personal protection and disinfection. Criteria were also established for upgrading or downgrading infection prevention strategies based on ARIMA results.ConclusionThese innovative methods, along with the ARIMA model, showed efficient infection protection for healthcare workers in close contact with COVID-19 infected patients, saving nearly 41% of the cost of maintaining high-level infection prevention measures and enhancing control of respiratory infections.

Time-varying parameters from the same period in previous years to improve runoff forecasting

The time-varying parameter functional (TVPF) method has shown promise in improving runoff forecasting under climate change and human activities. However, it has three notable limitations: data availability, increased model complexity, and the inability to update with newly measured data. To address these challenges, this study proposes an improved runoff forecasting method that relies solely on time-varying parameters from the same period in previous years. The proposed method comprises four steps: (1) identification of time-varying parameters from historical observations; (2) estimation of the current-year parameters by averaging the time-varying parameters from the same period in previous years; (3) runoff forecasts using the estimated current-year parameters; (4) continuous identification and estimation of parameters with newly measured data for each year, enabling a rolling forecast of runoff. The proposed method is compared with both the constant parameter method and TVPF method, using the Xun River and Zuo River basins as case studies. The results reveal that (1) parameters from the same period in the previous three years yield the best runoff forecasts in the Xun River basin, while parameters from the same period in the previous five years produce the best runoff forecasts in the Zuo River basin; (2) the proposed method significantly improves the water balance index (WBI) compared to the TVPF method, providing either superior or comparable performance in structural risk minimization (SRM) evaluation. Overall, this study provides an approach to forecast runoff in a changing environment that reduces data requirements, simplifies model complexity, and improves WBI performance.

Dynamic real-time forecasting technique for reclaimed water volumes in urban river environmental management

In recent years, the utilization of wastewater recycling as an alternative water source has gained significant traction in addressing urban water shortages. Accurate prediction of wastewater quantity is paramount for effective urban river water resource management. There is an urgent need to develop advanced forecasting technologies to further enhance the accuracy and efficiency of water quantity predictions. Decomposition ensemble models have shown excellent predictive capabilities but are challenged by boundary effects when decomposing the original data sequence. To address this, a rolling forecast decomposition ensemble scheme was developed. It involves using a machine learning (ML) model for prediction and progressively integrating prediction outcomes into the original sequence using complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN). Long short-term memory (LSTM) is then applied for sub-signal prediction and ensemble. The ML-CEEMDAN-LSTM model was introduced for wastewater quantity prediction, compared with non-decomposed ML models, CEEMDAN-based LSTM models, and ML-CEEMDAN-based LSTM models. Three ML algorithms—linear regression (LR), gradient boosting regression (GBR), and LSTM—were examined, using real-time prediction data and historical monitoring data, with historical data selected using the decision tree method. The study used daily water volumes data from two reclaimed water plants, CH and WQ, in Beijing. The results indicate that (1) ML models varied in their selection of real-time factors, with LR performing best among ML models during testing; (2) the ML-CEEMDAN-LSTM model consistently outperformed ML models; (3) the ML-CEEMDAN-LSTM hybrid model performed better than the CEEMDAN-LSTM model across different seasons. This study offers a reliable and accurate approach for reclaimed water volumes forecasting, critical for effective water environment management.

Ultra-Short-Term Load Forecast of Charging Stations Considering the Travelling and Charging Behaviours of Electric Vehicles

The traditional electric vehicle (EV) load forecasting methods are mostly used to predict large areas such as cities, which makes it difficult to serve practical applications. The existing charging station load forecasting methods ignore the traveling and charging behaviors of EVs. To this end, a charging station ultra-short-term load prediction method considering the traveling and charging behaviors of EVs is proposed. Firstly, EV types are classified according to charging behaviors. On this basis, a rolling forecast model of the number of arrivals per unit time of each vehicle type is established, and then the impact of charging randomness on load forecast is reduced by repeated Monte Carlo sampling. The total load of charging stations during the prediction period can be obtained by adding up the load of individual vehicles. The results show that the method enables accurate prediction of ultra-short-term load at charging stations, which has higher accuracy and reference value than the traditional load forecasting methods.

An upscaling minute-level regional photovoltaic power forecasting scheme

Along with the increasing penetration of photovoltaic (PV) power generation, regional power forecasting becomes more and more critical for stable and economical operation of power system. The key challenge of regional PV power forecasting technology is the lack of complete and accurate historical power data since not all PV plants are equipped with the precise real-time output power monitoring system. Besides, the computation burden will be heavy when the number of PV plants in the target region is large. This paper therefore proposes an upscaling minute-level regional PV power forecasting scheme using the data of the selected reference PV plants. In this paper, a novel method of reference PV plants selection is proposed by comprehensively considering the prediction accuracy of artificial neural network (ANN) as well as Pearson correlation coefficient. The reference PV plant selection coefficient μ is introduced as the comprehensive indicator for reference PV plant selection, which incorporates Pearson correlation coefficient and MAPE. In addition, a PV output power correction method is assumed to guarantee the proper operation of regional power forecasting. Besides, this paper proposes a flexible approach to effectively decrease the accumulated error of rolling forecasting by integrating the forecasting results under different temporal resolutions. In specific, the power forecasting results in temporal resolutions of 1 min, 5 min and 15 min are simultaneously derived and the performance between the traditional rolling forecasting and the proposed method is compared. The validity of the proposed method is finally verified using the collected historical power data of PV plants installed in a city of Eastern China. For time resolution of 1 min, 5 mins and 10 mins, the corresponding RMSE are 6.56, 5.73 and 4.85 and corresponding MAPE are 4.04%, 3.45% and 2.86%.

BiLSTM for Predicting Post-Construction Subsoil Settlement under Embankment: Advancing Sustainable Infrastructure

The load and settlement histories of stage-constructed embankments provide critical insights into long-term surface behavior under embankment loading. However, these data often remain underutilized in predicting post-construction settlement in the absence of geotechnical subsoil characterization. To address this limitation, the current study integrates bidirectional long short-term memory (BiLSTM) into a three-phase framework: data preparation, model construction, and performance evaluation. In the data preparation phase, the feature vector comprises basal pressure, pressure increments, time intervals, and prior settlement values to facilitate a rolling forecast. To manage unevenly spaced data, an Akima spline standardizes the desired time intervals. The model’s efficacy is validated using observational data from two distinct construction case studies, each featuring diverse soil conditions. BiLSTM proves effective in identifying key attributes from load and settlement data during the staged construction process. Compared to traditional curve-fitting methods, the BiLSTM model exhibits superior performance, robustness, and adaptability to varying soil conditions. Additionally, the model demonstrates low sensitivity to the range of post-construction data, allowing for a data collection period reduction—from six months to three—without compromising prediction accuracy (relative error = 0.92%). These advantages not only optimize resource allocation but also contribute to broader sustainability objectives.

Stochastic search for a parametric cost function approximation: Energy storage with rolling forecasts

Rolling forecasts have been almost overlooked in the renewable energy storage literature. In this paper, we provide a new approach for handling uncertainty not just in the accuracy of a forecast, but in the evolution of forecasts over time. Our approach shifts the focus from modeling the uncertainty in a lookahead model to accurate simulations in a stochastic base model. We develop a robust policy for making energy storage decisions by creating a parametrically modified lookahead model, where the parameters are tuned in the stochastic base model. Since computing unbiased stochastic gradients with respect to the parameters require restrictive assumptions, we propose a simulation-based stochastic approximation algorithm based on numerical derivatives to optimize these parameters. While numerical derivatives, calculated based on the noisy function evaluations, provide biased gradient estimates, an online variance reduction technique built in the framework of our proposed algorithm, will enable us to control the accumulated bias errors and establish the finite-time rate of convergence of the algorithm. Our numerical experiments show the performance of this algorithm in finding policies outperforming the deterministic benchmark policy.

Predicting air quality using a quantitative forecasting model of PM2.5 micro-scale variation nested with wave spectrum analysis

This study focused on weather and environmental numerical prediction and public demand. It expanded the concept and technology growth points in new fields in terms of new tasks for major prediction services for “large-scale public events.” This is required for more advanced prediction and to improve the resolution, fineness, and accuracy of the prediction. This study explored the prediction theory and technical application of transient atmospheric aerosol pollution within an accuracy of an hour. The novelty of this study is as follows: ①Based on high-quality big data covering the Northern Hemisphere with high temporal resolution with an accuracy of 1 h, a quantitative theory of the “natural weather cycle” spectral analysis algorithm was developed. This study presented a quantitative forecast model that nests the “spectral analysis of atmospheric wave-like disturbance” in the westerly belt with the “transient characteristics” of micro-scale aerosols (PM2.5 concentration) in Beijing and North China. ②According to the nested model of this study, the wave-like oscillation (H′) of 500 hPa was positively correlated with the PLAM index and PM2.5 mass concentration during nested multi-“natural weather cycles.” The significance level exceeded 0.001. This study demonstrated the prediction abilities of early quantitative fine prediction theory and implementation in the context of air quality. The forecast service on 1 October 2022, for the opening of the CCP 20th National Congress (16 October), and during the conference was successfully presented in real time. The results of this study on hourly resolution high-precision air quality forecasting service showed that rolling forecasts can be continuously released both 1 month and 7–10 days in advance, and the nesting effect can constantly be updated. Forecasts were found to be consistent with reality. ③The nested mode method for atmospheric spectrum analysis and micro-scale aerosol (PM2.5) distribution provides quantitative analysis and a decision-making basis for business-oriented operations to address technical difficulties.

Short-Term Dynamic Ampacity Forecasting for the Fully Insulated Busbar Taped Joint

The prior research on dynamic thermal rating have not taken into account the joint that could potentially form a hot spot in the power system. This paper proposes a novel method for short-term dynamic ampacity forecasting of a fully insulated busbar taped joint. The first-order thermal network, in which the thermal parameters are identified through FEM thermal analysis, simplifies 3-D temperature field of the joint. The surface temperature of the busbar is used to extract environmental information based on the equivalence principle of boundary conditions. Time series modelling is employed to forecast short-term environmental changes over the course of the following hour. On this basis, a rolling forecasting of the dynamic ampacity can be realized. A dynamic thermal rating test was carried out to validate and improve the model. The measured joint conductor temperature is nearly but consistently below the temperature limit of 90°C during the whole test. This indicates that the dynamic thermal rating is successfully accomplished in the laboratory using the proposed algorithm, increasing the transmission capacity of the insulated busbar.

Temperature Prediction Based on STOA-SVR Rolling Adaptive Optimization Model

In this paper, a support vector regression (SVR) adaptive optimization rolling composite model with a sooty tern optimization algorithm (STOA) has been proposed for temperature prediction. Firstly, aiming at the problem that the algorithm tends to fall into the local optimum, the model introduces an adaptive Gauss–Cauchy mutation operator to effectively increase the population diversity and search space and uses the improved algorithm to optimize the key parameters of the SVR model, so that the SVR model can mine the linear and nonlinear information in the data well. Secondly, the rolling prediction is integrated into the SVR prediction model, and the real-time update and self-regulation principles are used to continuously update the prediction, which greatly improves the prediction accuracy. Finally, the optimized STOA-SVR rolling forecast model is used to predict the final temperature. In this study, the global mean temperature data set from 1880 to 2022 is used for empirical analysis, and a comparative experiment is set up to verify the accuracy of the model. The results show that compared with the seasonal autoregressive integrated moving average (SARIMA), feedforward neural network (FNN) and unoptimized STOA-SVR-LSTM, the prediction performance of the proposed model is better, and the root mean square error is reduced by 6.33–29.62%. The mean relative error is reduced by 2.74–47.27%; the goodness of fit increases by 4.67–19.94%. Finally, the global mean temperature is predicted to increase by about 0.4976 °C in the next 20 years, with an increase rate of 3.43%. The model proposed in this paper not only has a good prediction accuracy, but also can provide an effective reference for the development and formulation of meteorological policies in the future.

Multi-scale regional photovoltaic power generation forecasting method based on sequence coding reconstruction

This paper proposes a solution for the challenges faced in predicting the output of distributed solar power stations due to difficulties in obtaining and utilizing data. The proposed method is a multi-scale regional photovoltaic power generation forecasting approach that uses sequence coding reconstruction. Firstly, the reference sites were selected to enhance the data in time domain and frequency domain for the limited site dataset. Then, the forecasting model of sequence coding reconstruction is constructed based on convLSTM, and the attention mechanism is added in the decoding process, so that it can make full use of the spatio-temporal correlation characteristics of regional photovoltaic power and realize the rolling forecasting of regional photovoltaic power. The result shows that the proposed method can maintain a good multi-scale regional photovoltaic power forecasting effect under the condition of limited data amount and low data collection cost, the accuracy of the proposed model is increased compared with the naive forecasting method.

Supply Chain Demand Forecasting and Price Optimisation Models with Substitution Effect

Determining the optimal price of products is essential, as it plays a critical role in improving a company’s profitability and market competitiveness. This requires the ability to calculate customers’ demand in the Fast Moving Consumer Goods (FMCG) industry as various effects exist between multiple products within a product category. The substitution effect is one of the challenging effects at retail stores, as it requires investigating an exponential number of combinations of price changes and the availability of other products. This paper suggests a systematic price decision support tool for demand prediction and price optimise in online and stationary retailers considering the substitution effect. Two procedures reflecting the product price changes and the demand correlation structure are introduced for demand prediction and price optimisation models. First, the developed demand prediction procedure is carried out considering the combination of price changes of all products reflecting the effect of substitution. Time series and different well-known machine learning approaches with hyperparameter tuning and rolling forecasting methods are utilised to select each product’s best demand forecast. Demand forecast results are used as input in the price optimisation model. Second, the developed price optimisation procedure is a constraint programming problem based on a week time frame and a product category level aggregation and is capable of maximising profit out of the many price combinations. The results using real-world transaction data with 12 products and 4 discount rates demonstrate that including some business rules as constraints in the proposed price optimisation model reduces the number of price combinations from 11,274,924 to 19,440 and execution time from 129.59 to 25.831 min. The utilisation of the presented price optimisation support tool enables the supply chain managers to identify the optimal discount rate for individual products in a timely manner, resulting in a net profit increase.

Machine learning models for forecasting and estimation of business operations

Machine Learning (ML) systems are built to shift through large amounts of data. Applying ML in production settings allows for the collection of additional data that can be used to guide future decisions about the system's design. Since the late 1970s, academics have taken an interest in the field of financial predictions. The real business environment has neglected statistical methods in forecasting, despite highly sophisticated models and rising competence in econometrics and economics studies. Current research centres on implementing various algorithms to identify the variation in performance for each product, and it compares the time series models to one another to identify the better model. A basic forecast model can make reliable, fact-based sales projections, as suggested by the books on forecasting. The worth of the forecast model lies in its ability to simplify the arduous tasks of budgeting and rolling forecasting by providing an unbiased forecast upon which a comprehensive financial strategy can be based. In this research, we first look for appropriate machine learning algorithms that can be used to predict sales of truck components, and then we run experiments with the selected algorithms to make predictions about sales and assess how well they work. Business forecasting allows for the estimation of a wide variety of activities, each of which can be tailored to the individual requirements of the company. Here are a few examples of frequently estimated kinds of operations. Although it is well-known that certain algorithms, such as Simple Linear Regression, Gradient Boosting Regression, Support Vector Regression, and Random Forest Regression, outperform others, it has been demonstrated that Random Forest Regression is the most suitable algorithm. Based on the results of the experiments and the analysis, the Ridge regression algorithm was selected as the best algorithm to conduct the sales forecasting of truck components for the selected data.

Constructing the accurate forecast: an actor-network theory approach

PurposeThis paper empirically demonstrates the major organizational events of a rolling forecasting process and the roles of controllers therein. In particular, this study aims to investigate how the understanding of a “realistic forecast” is translated and questioned by various mediators in the rolling forecasting process and how it affects the quality of planning as the ultimate accuracy of forecasts is seen as important.Design/methodology/approachThis study follows an actor-network theory (ANT) approach and maps the key points of translation in the rolling forecasting process by inspecting the roles of mediators. This qualitative case study is based on interviews with controllers and managers involved in the forecasting process in a single manufacturing company.FindingsThe paper identified two episodes of translation in the forecasting process, in which the forecast partially stabilized to create room for managerial discussion and debate. The abilities of controllers to infiltrate various functional groups and calculative practices appeared to be one way to control the accuracy of forecasting, although this was built on a façade of neutrality.Originality/valuePrior literature identifies the aims of interactive planning processes as being to improve the quality of planning. The authors apply ANT to better understand the nature of mediators in constructing an entity called a “realistic rolling forecast”.

The applicability of machine learning algorithms in accounts receivables management

PurposeThe accurate prediction of incoming cash flows enables more effective cash management and allows firms to shape firms' planning based on forward-looking information. Although most firms are aware of the benefits of these forecasts, many still have difficulties identifying and implementing an appropriate prediction model. With the rise of machine learning algorithms, numerous new forecasting techniques have emerged. These new forecasting techniques are theoretically applicable for predicting customer payment behavior but have not yet been adequately investigated. This study aims to close this research gap by examining which machine learning algorithm is the most appropriate for predicting customer payment dates.Design/methodology/approachBy using various machine learning algorithms, the authors evaluate whether customer payment behavior patterns can be identified and predicted. The study is based on real-world transaction data from a DAX-40 firm with over 1,000,000 invoices in the dataset, with the data covering the period 2017–2019.FindingsThe authors' results show that neural networks in particular are suitable for predicting customers' payment dates. Furthermore, the authors demonstrate that contextual and logical prediction models can provide more accurate forecasts than conventional baseline models, such as linear and multivariate regression.Research limitations/implicationsFuture cash flow forecasting studies should incorporate naïve prediction models, as the authors demonstrate that these models can compete with conventional baseline models used in existing machine learning research. However, the authors expect that with more in-depth information about the customer (creditworthiness, accounting structure) the results can be even further improved.Practical implicationsThe knowledge of customers' future payment dates enables firms to change their perspective and move from reactive to proactive cash management. This shift leads to a more targeted dunning process.Originality/valueTo the best of the authors' knowledge, no study has yet been conducted that interprets the prediction of incoming payments as a daily rolling forecast by comparing naïve forecasts with forecasts based on machine learning and deep learning models.