Nazario D. Ramirez-Beltran

Application of neural networks to chemical process control

An artificial neural network is used to model and control the pH of the erythromycin acetate salt. Experiments were mainly conducted to determine the time delay of chemical reactions at the Abbott Chemical Plant located in Barceloneta, Puerto Rico. The suggested methodology includes three main steps: (1) the cross-correlation function is used to detect time delay, (2) a feedforward neural network is used to model the input and output variables of a nonlinear dynamic process, and (3) an optimization technique is used to solve the control equation and implement the corrective action. The selected neural network algorithm works as an adaptive procedure. The implemented algorithm reads the last 60 observations from four variables to generate a recommendation for controlling the pH of the erythromycin acetate salt.

Neural networks for on-line parameter change detections in time series models

Time series models can be derived directly from a manufacturing process, assuming that variables are observed at equal time intervals. A theoretical autocorrelation function for the identified model is developed based on the estimated parameters. The Monte Carlo simulation technique is used to generate a synthetic series which has a similar autocorrelation function to the theoretical one. A feedforward neural network is trained to recognize the patterns exhibited in a sample autocorrelation function. An on-line sequential window obtained from a continuous process is propagated into the network to detect parameter changes for a given manufacturing process.

Neural networks to model dynamic systems with time delays

Abstract An algorithm is proposed to identify a neural network model that represents a nonlinear dynamic system with a multivariate time delay response. The algorithm consists of two major parts. The first one identifies the time delay vector for a given neural network structure. This task is accomplished by using an exhaustive integer enumeration algorithm that minimizes a statistical parameter to assess the performance of the neural network model. The second part uses a cross-validation strategy to identify the best neural network model. Since the structure that models a nonlinear system is usually unknown, the identification strategy consists of selecting several neural network structures and identifying the best time delay vector for each network. The modeling process starts with the simplest structure and progressively the complexity of the network is increased to end up with a complex structure. Finally, the network that offers the simplest structure with the best network performance is the one that exhibits the appropriate neural network structure with the corresponding optimal time delay vector. The Monte Carlo simulation technique was used to test the performance of the algorithm under the presence of linear and nonlinear relationships among several variables of dynamic systems and with a different time delay applied to each input variable. The introduced algorithm is used to detect a chemical reaction delay among enriched amyl acetate, acetic acid, water, and the pH of erythromycin sail. An appropriate neural network model was designed to model the pH of the erythromycin during a continuous extraction process. To the best of the authors knowledge the proposed algorithm is the only one currently available to identify time delay interactions in the multivariate input output variables of a system. The major drawback of the introduced algorithm is that it becomes very slow as the number of system inputs increases. This algorithm works efficiently in a system that involves five inputs or less.

Empirical Probability Models to Predict Precipitation Levels over Puerto Rico Stations

Abstract A new algorithm is proposed to predict the level of rainfall (above normal, normal, and below normal) in Puerto Rico that relies on probability and empirical models. The algorithm includes a theoretical probability model in which parameters are expressed as regression equations containing observed meteorological variables. Six rainfall stations were used in this study to implement and assess the reliability of the models. The stations, located throughout Puerto Rico, have monthly records that extend back 101 yr. The maximum likelihood method is used to estimate the parameters of the empirical probability models. A variable selection (VS) algorithm identifies the minimum number of variables that maximize the correlation between predictors and a predictand. The VS algorithm is used to identify the initial point and the maximum likelihood is optimized by using the sequential quadratic programming algorithm. Ten years of cross validation were applied to the results from six stations. The proposed met...

An algorithm to estimate soil moisture over vegetated areas based on in situ and remote sensing information

An algorithm is proposed for estimating soil moisture over vegetated areas. The algorithm uses in situ and remote sensing information and statistical tools to estimate soil moisture at 1 km spatial resolution and at 20 cm depth over Puerto Rico. Soil moisture within the study region is characterized by spatial and temporal variability. The temporal variability for a given area exhibits long- and short-term variations that can be expressed by two empirical models. The average monthly soil moisture exhibits the long-term variability and is modelled by an artificial neural network (ANN), whereas the short-term variability is determined by hourly variation and is represented by a nonlinear stochastic transfer function model. Monthly vegetation index, land surface temperature, accumulated rainfall and soil texture are the major drivers of the ANN to estimate the monthly soil moisture. Radar, satellite and in situ observations are the major sources of information of the soil moisture empirical models. A self-organized ANN was also used to identify spatial variability to be able to determine a similar transfer function that best resembles the properties of a particular grid point and estimate the hourly soil moisture across the island. Validation techniques reveal an average absolute error of 3.34% of volumetric water content and this result shows that the proposed algorithm is a potential tool for estimating soil moisture over vegetated areas.

Transfer function models to control a chemical process

Abstract Transfer function models are used to express the relationship between the input and the output variables of a continuous chemical process. A time series method is used to derive the control equation. Experiments were conducted at a chemical plant to estimate the time delay and the impulse response function between the input and the output variables. The input variable that perturbs the chemical process is called the disturbance variable and the associated transfer function is called the perturbation equation. Similarly, the input variable that neutralizes the disturbance effect is called the compensation variable and its transfer function is called the compensation equation. Correlation analysis was used to identify the mathematical structure of the disturbance and compensation equations. Monte Carlo simulation technique and the Hook and Jeeves search were used to determine the initial parameter set for a nonlinear optimization routine. The Levenberg-Marquardt algorithm was used to obtain the final estimates of the transfer function. The control equation determines the values of the compensation variable that should be input into the system to neutralize the disturbance effects and maintain the output variable as close as possible to the target value.

Application of an heuristic procedure to solve mixed-integer programming problems

Abstract An heuristic algorithm is proposed to solve mixed integer programming problems. The optimal and suboptimal continuous solutions are first identified. Then, an integer solution is found in the neighborhood of each suboptimal and optimal point. If an integer point provides an infeasible solution, then the dual simplex method is used to derive a feasible integer solution. The suggested algorithm is derived under the framework of an integer exploratory search principle. Once an integer solution is found at each optimal and suboptimal point, the best point is called the heuristic solution for the underlying problem. The heuristic algorithm has successfully been applied to solve a production planning problem and it is compared with a well known commercial computer package,CPLEX.

Analysis of the Heat Index in the Mesoamerica and Caribbean Region

AbstractHourly data collected from ground stations were used to study the maximum daytime heat index Hi in the Mesoamerica and Caribbean Sea (MAC) region for a 35-yr period (1980–2014). Observations of Hi revealed larger values during the rainy season and smaller values during the dry season. The Hi climatology exhibits the largest values in Mesoamerica, followed by the Greater Antilles and then by the Lesser Antilles. The trend in Hi indicates a notable increasing pattern of 0.05°C yr−1 (0.10°F yr−1), and the trends are more prominent in Mesoamerica than in Caribbean countries. This work also includes the analysis of heat index extreme events (HIEE). Usually the extreme values of the heat index are used for advising heat warning events, and it was found that 45 HIEEs occurred during the studied period. The average duration of HIEE was 2.4 days, and the average relative intensity (excess over the threshold) was 2.4°C (4.3°F). It was found that 82% of HIEE lasted 2 or 2.5 days and 80% exhibited relative in...

A vector autoregressive model to predict hurricane tracks

The stochastic behaviour of hurricane tracks is expressed by a vector autoregressive time series model. Historical data and correlation analysis were used to identify the model structure of a typical hurricane track. The parameter estimation scheme is based on recursive and iterative algorithms. The recursive approach is used when a small number of points have been collected from a hurricane. On the Other hand, iterative algorithms are used when enough information for optimal estimation is available. The multivariate time series model was used to predict hurricane tracks during the 1990 hurricane season in the North Atlantic ocean. Prediction errors from the vector autoregressive model are compared with errors from the NHC90 model. The NHC90 model performs better than the studied model; however, the vector autoregressive model uses a small amount of information and may help to reduce official forecasting errors.

Projections of Heat Waves Events in the Intra-Americas Region Using Multimodel Ensemble

Significant accelerated warming of the Sea Surface Temperature of 0.15°C per decade (1982–2012) was recently detected, which motivated the research for the present consequences and future projections on the heat index and heat waves in the intra-Americas region. Present records every six hours are retrieved from NCEP reanalysis (1948–2015) to calculate heat waves changes. Heat index intensification has been detected in the region since 1998 and driven by surface pressure changes, sinking air enhancement, and warm/weaker cold advection. This regional warmer atmosphere leads to heat waves intensification with changes in both frequency and maximum amplitude distribution. Future projections using a multimodel ensemble mean for five global circulation models were used to project heat waves in the future under two scenarios: RCP4.5 and RCP8.5. Massive heat waves events were projected at the end of the 21st century, particularly in the RCP8.5 scenario. Consequently, the regional climate change in the current time and in the future will require special attention to mitigate the more intense and frequent heat waves impacts on human health, countries’ economies, and energy demands in the IAR.