Adaptive Prediction Model Research Articles

A novel clustering-enhanced adaptive artificial neural network model for predicting day-ahead building cooling demand

To accurately predict hourly day-ahead building cooling demand, year-round historical weather profile needs to be evaluated. The daily weather profiles among different time periods result in various features of historical datasets. The different appropriate structure and parameters of artificial neural network models may be identified for training datasets with different features. In this study, a novel clustering-enhanced adaptive artificial neural network (C-ANN) model is proposed to forecast 24h-ahead building cooling demand in subtropical areas. The uniqueness of the proposed adaptive model is that k-means clustering is implemented to recognise representative patterns of daily weather profile and thus categorize the annual datasets into featuring clusters. Each cluster of the weather profile, along with the corresponding time variables and cooling demand, is adopted to train one ANN sub-model. The optimal structure and parameters of each ANN sub-model are selected according to its featuring training datasets; thus the ANN sub-models are adaptive. The proposed C-ANN model is tested on a representative office building in Hong Kong. It is found that the mean absolute percentage error of the training and testing cases of the proposed predictive model is 3.59% and 4.71%, which has 4.2% and 3.1% improvement compared to conventional ANN model with a fixed structure. The proposed adaptive predictive model can be applied in building energy management system to accurately predict day-ahead building cooling demand using the latest forecast weather profile.

Implementation Strategy for Resource Optimization of FPGA-Based Adaptive Finite Control Set-MPC Using XSG for a VSI System

Finite control set-model predictive control (FCS-MPC) for the power electronic systems has been implemented using the field-programmable gate array (FPGA), considering an alternative solution to handle the computational burden of the control algorithm. However, FPGA resource utilization is a concern for optimum design and development of the FCS-MPC. This article presents analytical approach-based implementation strategies using the Xilinx system generator (XSG), considering the real-time operation of the FCS-MPC for a three-phase voltage source inverter (VSI). Moreover, an adaptive prediction model is proposed to adapt to the change in the load current. The percentage total harmonic distortion (THD), dynamic response, and steady-state error are evaluated for system-performance analysis. The implementation strategies are compared based on FPGA resource utilization for the FCS-MPC in the stationary αβ- and rotating dq-frames considering the fixed, approximated, and adaptive values of a coefficient used for current prediction. The lookup tables (LUTs), flip-flops, and digital signal processor (DSP) slices have been opted as the utilization indices of the FPGA resources for comparative analysis. The optimum design-based controller model is used for FPGA-based experimental system implementation. The simulation and experimental results are used for the performance evaluation of the FCS-MPC in both αβ- and dq -frames. Moreover, change in the switching-state constraint is included in the cost function and controller performance is evaluated corresponding to the weighting factor for the approximated and adaptive coefficient values.

Adaptive Modeling Strategy Integrating Feature Selection and Random Forest for Fluid Catalytic Cracking Processes

This study proposes a hybrid approach for the modeling of the fluid catalytic cracking (FCC) process, with the aim to establish an adaptive and accurate product yield prediction model. Because of t...

Adaptive neuro fuzzy predictive models of agricultural biomass standard entropy and chemical exergy based on principal component analysis

In order to effectively utilize energy of agricultural biomass, there is a need to evaluate energy potential. For such a purpose, chemical exergy and standard entropy of typical agricultural biomass were examined analytically. Element compositions of the exergy and entropy were acquired for further statistical evaluation. Adaptive neuro fuzzy inference system (ANFIS) was used as the statistical methodology for data analyzing. ANFIS is an efficient estimation model among machine learning techniques. The main weakness of the ANFIS is its dimensionality problem with large inputs. Therefore, the main goal in this study was to estimate the parameters’ influence on the chemical exergy and standard entropy prediction in order to reduce the number of inputs. Principal component analysis was used for presentation of the obtained ANFIS predictive models. Obtained results have shown the best predictive performances for standard entropy based on hydrogen as composite element of the agricultural biomass. Exergy prediction was the best for oxygen as composite element of the agricultural biomass. ANFIS coefficient of determination for standard entropy prediction based on hydrogen is 0.9832 and for chemical exergy prediction is 0.919. The results show the high predictive accuracy of ANFIS models.

Endpoint forecast of different diesel-biodiesel soot filtration process in diesel particulate filters considering ash deposition

In order to effectively forecast endpoint of soot loading process in the diesel particulate filter (DPF), an efficient prediction method is presented in this work. Firstly, ash deposition mass is predicted by the fuzzy adaptive variable weight functional link neural network model. Then, pressure drop of the DPF is simulated by a modified soot filtration mathematical model considering ash deposition. Finally, the soot loading endpoints with different fuels and initial ash mass are forecasted based on the cusp catastrophe model. The results show that the fuzzy adaptive variable weight functional link neural network prediction model has higher prediction accuracy with 2.24% average error than other single prediction methods. In addition, pressure drop variation rate of the DPF increases over time and it obviously rises with the increase of the pre-loaded ash mass, DPF with larger initial ash mass has a shorter soot loading time to reach the same pressure drop, and soot mass decreases with the rise of biodiesel proportion in the fuels at the same moment. Moreover, predicted pressure drop and discriminant value Δ indicate that a DPF with elevated ash loads has shorter soot loading time and lower soot mass, biodiesel or its blends can prolong the soot filtration time and the optimal range of endpoint time during B0 soot loading process is between 4.25 h and 4.5 h for a clean DPF. This work offers us great reference value for forecasting soot loading endpoint and managing regeneration of the periodic regenerated particulate filters.

Predicting Adverse Outcomes for Febrile Patients in the Emergency Department Using Sparse Laboratory Data: Development of a Time Adaptive Model.

BackgroundA timely decision in the initial stages for patients with an acute illness is important. However, only a few studies have determined the prognosis of patients based on insufficient laboratory data during the initial stages of treatment.ObjectiveThis study aimed to develop and validate time adaptive prediction models to predict the severity of illness in the emergency department (ED) using highly sparse laboratory test data (test order status and test results) and a machine learning approach.MethodsThis retrospective study used ED data from a tertiary academic hospital in Seoul, Korea. Two different models were developed based on laboratory test data: order status only (OSO) and order status and results (OSR) models. A binary composite adverse outcome was used, including mortality or hospitalization in the intensive care unit. Both models were evaluated using various performance criteria, including the area under the receiver operating characteristic curve (AUC) and balanced accuracy (BA). Clinical usefulness was examined by determining the positive likelihood ratio (PLR) and negative likelihood ratio (NLR).ResultsOf 9491 eligible patients in the ED (mean age, 55.2 years, SD 17.7 years; 4839/9491, 51.0% women), the model development cohort and validation cohort included 6645 and 2846 patients, respectively. The OSR model generally exhibited better performance (AUC=0.88, BA=0.81) than the OSO model (AUC=0.80, BA=0.74). The OSR model was more informative than the OSO model to predict patients at low or high risk of adverse outcomes (P<.001 for differences in both PLR and NLR).ConclusionsEarly-stage adverse outcomes for febrile patients could be predicted using machine learning models of highly sparse data including test order status and laboratory test results. This prediction tool could help medical professionals who are simultaneously treating the same patient share information, lead dynamic communication, and consequently prevent medical errors.

Monthly Mean Meteorological Temperature Prediction Based on VMD-DSE and Volterra Adaptive Model

Climate is a complex and chaotic system, and temperature prediction is a challenging problem. Accurate temperature prediction is also concerned in the fields of energy, environment, industry, and agriculture. In order to improve the accuracy of monthly mean temperature prediction and reduce the calculation scale of hybrid prediction process, a combined prediction model based on variational mode decomposition-differential symbolic entropy (VMD-DSE) and Volterra is proposed. Firstly, the original monthly mean meteorological temperature sequence is decomposed into finite mode components by VMD. The DSE is used to analyze the complexity and reconstruct the sequences. Then, the new sequence is reconstructed in phase space. The delay time and embedding dimension are determined by the mutual information method and G-P method, respectively. On this basis, the Volterra adaptive prediction model is established to modeling and predicting each component. Finally, the final predicted values are obtained by superimposing the predicted results. The monthly mean temperature data of Xianyang and Yan’an are used to verify the prediction performance of the proposed model. The experimental results show that the VMD-DSE-Volterra model shows better performance in the prediction of monthly mean temperature compared with other benchmark models in this paper. In addition, the combined forecasting model proposed in this paper can reduce the modeling time and improve the forecasting accuracy, so it is an effective forecasting model.

Analysis on the construction of sports match prediction model using neural network

To grasp the development of sports events in time and adjust the strategy in the process of match in time, the traditional back-propagation neural network (BPNN) algorithm was improved, and the match prediction model was constructed by using the adaptive BPNN. Taking the football match data of the Union of European Football Associations Champions League 2016–2017 as the prediction sample, the match was predicted. Moreover, taking some match data of Barcelona in 2016–2017 as an example, the fitting accuracy of the improved adaptive BPNN prediction model, multiple linear regression (MLR) model and grey degree prediction model were compared and analyzed. The research results showed that the prediction model built by the improved adaptive BPNN algorithm had smaller prediction error after rolling prediction. By comparing with the fitting accuracy of MLR model and grey prediction model, it was also found that the prediction error of the prediction model proposed was almost zero, and the error of the other two models was large. It showed that the prediction model proposed had high accuracy and reliability. Therefore, applying neural network to build a sports match prediction model and then predict the match results can provide a certain theoretical basis for sports match practice and the prediction and analysis of the match results.

GWLM–NARX

PurposeWeather forecasting is the trending topic around the world as it is the way to predict the threats posed by extreme rainfall conditions that lead to damage the human life and properties. These issues can be managed only when the occurrence of the worse weather is predicted in advance, and sufficient warnings can be executed in time. Thus, keeping in mind the importance of the rainfall prediction system, the purpose of this paper is to propose an effective rainfall prediction model using the nonlinear auto-regressive with external input (NARX) model.Design/methodology/approachThe paper proposes a rainfall prediction model using the time-series prediction that is enabled using the NARX model. The time-series prediction ensures the effective prediction of the rainfall in a particular area or the locality based on the rainfall data in the previous term or month or year. The proposed NARX model serves as an adaptive prediction model, for which the rainfall data of the previous period is the input, and the optimal computation is based on the proposed algorithm. The adaptive prediction using the proposed algorithm is exhibited in the NARX, and the proposed algorithm is developed based on the Grey Wolf Optimization and the Levenberg–Marqueret (LM) algorithm. The proposed algorithm inherits the advantages of both the algorithms with better computational time and accuracy.FindingsThe analysis using two databases enables the better understanding of the proposed rainfall detection methods and proves the effectiveness of the proposed prediction method. The effectiveness of the proposed method is enhanced and the accuracy is found to be better compared with the other existing methods and the mean square error and percentage root mean square difference of the proposed method are found to be around 0.0093 and 0.207.Originality/valueThe rainfall prediction is enabled adaptively using the proposed Grey Wolf Levenberg–Marquardt (GWLM)-based NARX, wherein an algorithm, named GWLM, is proposed by the integration of Grey Wolf Optimizer and LM algorithm.

Chaotic time series prediction using wavelet transform and multi-model hybrid method

In order to further improve the prediction accuracy of the chaotic time series and overcome the defects of the single model, a multi-model hybrid model of chaotic time series is proposed. First, the Discrete Wavelet Transform (DWT) is used to decompose the data and obtain the approximate coefficients (low-frequency sequence) and detailed coefficients (high-frequency sequence) of the sequence. Secondly, phase space reconstruction is performed on the decomposed data. Thirdly, the chaotic characteristics of each sequence are judged by correlation integral and Kolmogorov entropy. Fourthly, in order to explore the deeper features of the time series and improve the prediction accuracy, a sequence of Volterra adaptive prediction models is established for the components with chaotic characteristics according to the different characteristics of each component. For the components without chaotic characteristics, a JGPC prediction model without chaotic feature sequences is established. Finally, the multi-model fusion prediction of the above multiple sequences is carried out by the LSTM algorithm, and the final prediction result is obtained through calculation, which further improves the prediction accuracy. Experiments show that the multi-model hybrid method of Volterra-JGPC-LSTM is more accurate than other comparable models in predicting chaotic time series.

Distributed Joint Cooperative Self-Localization and Target Tracking Algorithm for Mobile Networks.

Location information is a key issue for applications of the Internet of Things. In this paper, we focus on mobile wireless networks with moving agents and targets. The positioning process is divided into two phases based on the factor graph, i.e., a prediction phase and a joint self-location and tracking phase. In the prediction phase, we develop an adaptive prediction model by exploiting the correlation of trajectories within a short period to formulate the prediction message. In the joint positioning phase, agents calculate the cooperative messages according to variational message passing and locate themselves. Simultaneously, the average consensus algorithm is employed to realize distributed target tracking. The simulation results show that the proposed prediction model is adaptive to the random movement of nodes. The performance of the proposed joint self-location and tracking algorithm is better than the separate cooperative self-localization and tracking algorithms.

Comparative thermal comfort study in educational buildings in autumn and winter seasons

Maintaining thermal comfort throughout an entire year is complicated because conditions vary greatly throughout the seasons. The aim of this article was to understand the influence of seasonal characteristics on the thermal comfort in classrooms and to propose a method to select suitable adaptive coefficients for different conditions. Field investigations were conducted in an educational building in Tianjin during autumn and winter. In the study, multiple environmental parameters were investigated, including air temperature, black globe temperature, relative humidity, air velocity, and the vertical temperature difference (VTD). The respondents’ thermal sensations were collected via questionnaires. The respondents’ thermal sensation and the influence of seasonal characteristics on thermal sensation were analyzed. The results showed that the thermal neutral temperature varied in different seasons, and the deviations between the mean thermal sensation vote (MTSV) and the predicted mean vote (PMV) showed obvious discrepancies throughout the different seasons. Furthermore, the adaptive coefficient αs for the adaptive thermal comfort prediction model (APMV) were 0.25 and 0.57 for autumn and winter, respectively. Based on the aforementioned result, a discriminant function (DFA) using the seasonal factors to select the suitable α was obtained.

Choosing the optimal hybrid covariance estimators in adaptive elastic net regression models using information complexity

ABSTRACTElastic net (EN) is a regularization technique which is used for modelling and variable selection with high-dimensional data at the same time. In the literature, it is claimed that EN modelling can be used for undersized samples with high dimensions (i.e. ). But, both the model matrix and the Gram matrix are not of full rank and the inverse of the Gram matrix cannot be calculated. It degenerates and becomes singular. To overcome this problem in EN modelling, Mohebbi et. al. [A new data adaptive elastic net predictive model using hybridized smoothed covariance estimators with information complexity. J Stat Comput Simul. 2019;89(6):1060–1089] purposed a new adaptive elastic net (AEN) modelling using hybrid covariance estimators (HCEs) and information complexity (ICOMP) criteria. There are several forms of HCEs which can be used in AEN regression modelling. Thus, how to decide which HCEs is appropriate is an important problem to be solved. In this paper, we study the performance of the AEN models under several different HCEs using the ICOMP criterion for both the implementation of experimental data and Monte Carlo simulation study with different scenarios of the protocol.

Runoff Prediction Method Based on Adaptive Elman Neural Network

The prediction of medium- and long-term runoff is of great significance to the comprehensive utilization of water resources. Building an adaptive data-driven runoff prediction model by automatic identification of multivariate time series change in runoff forecasting and identifying its influence degree is an attractive and intricate task. At present, the commonly used screening factor method is correlational analysis; others offer multi-collinearity. If these factors are directly input into the model, the parameters of the model tend to increase, and the excessive redundancy and noise adversely affects the prediction results of the model. On the basis of previous studies on medium- and long-term runoff prediction methods, this paper proposes an Elman Neural Network (ENN) adaptive runoff prediction method based on normalized mutual information (NMI) and kernel principal component analysis (KPCA). In this method, the features of the screening factors are extracted automatically by using the mutual information automatic screening factor, and then input into the Elman Neural Network for training. With less features, the parameters of the Elman Neural Network model can be reduced, and the problem of overfitting of the Elman Neural Network model is effectively alleviated. The method is evaluated by using the annual average runoff data of Jinping hydropower station in Chengdu, China, from 2007 to 2011. The maximum relative error of multiple forecasts was found to be less than 16%, and forecast effect was good. The accuracy of prediction is further improved by averaging the results of multiple forecasts.

Adaptive prediction model for effective electrical machine maintenance

Purpose The purpose of this paper is to propose an optimal predictive model for the short-term forecast of real-time non-stationary machine variables by combining time series prediction with adaptive algorithms to minimize the error and to improve the prediction accuracy. Design/methodology/approach The proposed model is applied for prediction of speed and controller set point of three-phase induction motor operating on closed loop speed control with AC drive and PI controller. At Stage 1, the trend of the machine variables has been extracted and added to auto-regressive moving average (ARMA) time series prediction. ARMA prediction has been carried out using different combinations of AR and MA methods in order to make prediction with less Mean Squared Error (MSE). Findings The prediction error indicates the inadequacy of the model to estimate the data characteristics, which has been resolved at the subsequent stage by cascading an adaptive least mean square finite impulse response filter to the time series model. The adaptive filter receives the predicted output including training data and iteratively adjusts its coefficients for zero error convergence. Research limitations/implications The componentized data prediction based on time series and cascade adaptive filter algorithm decomposes the non-stationary data characteristics for predictive maintenance. Evaluation of the model with different combination of time series algorithms and parameter settings of adaptive filter has been carried out to illustrate the performance of the prediction model. This prediction accuracy is compared with existing linear adaptive filter prediction using MSE as comparison index. The wide margin in the MSE values substantiates the prediction efficiency of the proposed model for machine data. Originality/value This model predicts the dynamic machine data with component decomposition at high accuracy, which enables to interpret the system response under dynamic conditions efficiently.

An Approach to a Clock Offsets Model for Real-Time PPP Time and Frequency Transfer During Data Discontinuity

To resolve the dilemma in any post-processing strategy, i.e., the difficulty of monitoring the real-time time and frequency signals in a timely manner, real-time GPS time and the frequency transfer have recently become trending topics. Unfortunately, data interruption occurs when conducting real-time time transfer, sometimes from unexpected reasons. In this study, to ensure the stability and precision of real-time time transfer, an adaptive prediction model and a between-epoch constraint receiver clock model are applied as the mathematic models. The purpose of prediction is to solve the ambiguity from re-convergence when the data reappear. Moreover, compared to the conventional method, the between-epoch constraint receiver clock model is employed in this study to consider the correlation of epoch-wise clock parameters to avoid wasting useful information. The simulation data and real data are compared to verify the performance of the new approach. The simulation data for 165 days are designed with random daily interruptions of 10, 30, 60 and 90 min. Real data from 12 days is captured from the incomplete data in routine observation records. Ignoring the simulation data and real data, the investigation of six stations shows that the results with the between-epoch constraint receiver clock model were smoother than those with a white noise model. With an adaptive prediction model and the between-epoch constraint receiver clock model, the simulation results illustrate that the average root mean squares (RMS) values of all the stations are significantly reduced, i.e., by 66.03% from 0.43 to 0.14 ns, by 64.91% from 0.44 to 0.15 ns, by 57.47% from 0.43 to 0.18 ns, and by 51.67% from 0.44 to 0.21 ns for the 10, 30, 60 and 90 min data interruptions, respectively. The stability of all the stations is improved by at least 50%. The improvement increases to 100% for short-term stability. The real results show that the stability of four links is boosted by at least 5%. The model proposed in this paper is more effective in producing short-term stability than long-term stability.

An Early Feedback Prediction System for Learners At-Risk Within a First-Year Higher Education Course

Identifying at-risk students as soon as possible is a challenge in educational institutions. Decreasing the time lag between identification and real at-risk state may significantly reduce the risk of failure or disengage. In small courses, their identification is relatively easy, but it is impractical on larger ones. Current Learning Management Systems store a large amount of data that could help to generate predictive models to early identification of students in online and blended learning. The contribution of this paper is twofold: First, a new adaptive predictive model is presented based only on students’ grades specifically trained for each course. A deep analysis is performed in the whole institution to evaluate its performance accuracy. Second, an early warning system is developed, focusing on dashboards visualization for stakeholders (i.e., students and teachers) and an early feedback prediction system to intervene in the case of at-risk identification. The early warning system has been evaluated in a case study on a first-year undergraduate course in computer science. We show the accuracy of the correct identification of at-risk students, the students’ appraisal, and the most common factors that lead to at-risk level.

A new data adaptive elastic net predictive model using hybridized smoothed covariance estimators with information complexity

ABSTRACTWe develop a novel Adaptive Elastic Net (AEN) modelling using a new covariance-regularization approach via the Hybridized Smoothed Covariance Estimators (HSCEs) to identify and select the best subset of predictors in undersized high-dimensional data sets. We introduce and score the Consistent and Misspecification Resistant Information Measure of Complexity criterion, and the Extended Consistent Akaike's Information Criterion with Fisher Information in AEN models for the first time. We carry out a large Monte Carlo simulation study using the medianmean-squared-error (MMSE) to demonstrate and compare the performance of the MMSE prediction. This is done using Cross-validated Fit Adaptive Elastic Net (CV-AEN) to avoid double shrinkage by varying both the error variance and the correlation structure of the model. Later, the new proposed AEN model is applied to a real undersized benchmark data set to predict the Riboflavin (Vitamin B2) production to select the best subset of predictors to predict the production rate of vitamin B2 and provide the best predictive model. The proposed new approach enables a simple and reliable identification of the best subset of predictive genes of the production rate of Riboflavin (Vitamin B2) without an exhaustive search of all possible subset selection in undersized high-dimensional data. It is a new and novel approach that has generalizability to other regularized General Linear Regression (GLM) models to determine the best predictor space for undersized data.

Occupancy prediction model for open-plan offices using real-time location system and inhomogeneous Markov chain

Implementing intelligent control strategies of building systems can significantly improve building energy performance and maintain or increase occupants' comfort level. However, these control strategies are dependent on the occupancy models. A good occupancy prediction model requires enough input data pertinent to the occupants' space utilization patterns. Nevertheless, most of the occupancy detection systems cannot provide this detailed information. As a result, most of the research works that considered shared multi-occupied offices did not distinguish between different individuals. Therefore, their practicality is reduced when they are used for open-plan offices. In this study, the occupancy modeling (i.e., occupants’ profiles) has been further enhanced using inhomogeneous Markov chain prediction model based on real occupancy data collected by a Real Time Locating System (RTLS). After extracting the detailed occupancy information with varying time-steps from the collected RTLS occupancy data, an adaptive probabilistic occupancy prediction model is developed. The comparison between the occupancy profiles resulting from the prediction model and the actual profiles showed that the prediction model was able to capture the actual behavior of occupants at occupant and zone levels with high accuracy. The proposed model distinguishes the temporal behavior of different occupants within an open-plan office and can be used for various levels of resolution required for the application of intelligent, occupancy-centered local control strategies of different building systems. This would eventually lead to a more robust control of building systems as well as more satisfied occupants.

An Approach of Anomaly Detection and Neural Network Classifiers to Measure Cellulolytic Activity.

A common method used for massive detection of cellulolytic microorganisms is based on the formation of halos on solid medium. However, this is a subjective method and real-time monitoring is not possible. The objective of this work was to develop a method of computational analysis of the visual patterns created by cellulolytic activity through artificial neural networks description. Our method learns by an adaptive prediction model and automatically determines when enzymatic activity on a chromogenic indicator such as the hydrolysis halo occurs. To achieve this goal, we generated a data library with absorbance readings and RGB values of enzymatic hydrolysis, obtained by spectrophotometry and a prototype camera-based equipment (Enzyme Vision), respectively. We used the first part of the library to generate a linear regression model, which was able to predict theoretical absorbances using the RGB color patterns, which agreed with values obtained by spectrophotometry. The second part was used to train, validate, and test the neural network model in order to predict cellulolytic activity based on color patterns. As a result of our model, we were able to establish six new descriptors useful for the prediction of the temporal changes in the enzymatic activity. Finally, our model was evaluated on one halo from cellulolytic microorganisms, achieving the regional classification of the generated halo in three of the six classes learned by our model. We assume that our approach can be a viable alternative for high throughput screening of enzymatic activity in real time.