Matemáticas aplicadas, estadística y computación

 

Role of automation agents in agribusiness decision support systems

 

Función de los agentes de automatización en los sistemas de apoyo a la toma de decisiones en agronegocios

 

Martin Pavlovic^1*, Fotis N. Koumboulis², Maria P. Tzamtzi² and Crtomir Rozman³

 

¹ International Hop Growers' Convention - Secretary General, 22, Rue des roses, F-67173 Brumath, France. *Author for correspondence. (martin.pavlovic@guest.arnes.si).

² Department of Automation, Halkis Institute of Technology, Greece.

³ Faculty of Agriculture, University of Maribor, Slovenia.

 

Received: August, 2007.
Aproved: September, 2008.

 

Abstract

Various automation agents can be embedded in a modern Decision Support System in agriculture, based on the following design approaches: E-Learning Agent, Monitoring Agent, Fault Diagnosis Agent, Weather Information Agent, E-Commerce Agent and Logistics Agent. The proposed Agents contribute to farmers' education, to timely correct fault diagnosis and the incorporation of modern agricultural techniques. Moreover, they allow processing historical data, while taking into account several factors in order to support decision making. The preliminary Automation Agents for Decision Support Systems (AADSS) model idea may be applied to various agribusiness activities on international level. Moreover, an operator agent is proposed, that exploits technological and economic information gathered from all the aforementioned Automation Agents in order to make decisions regarding the proposal of possible scenarios in a chain of raw material production to a final product. An initiative to develop and introduce stepwise a model for hop industry decision support system with embedded automation agents has been launched within the International Hop Growers' Convention (IHGC). It is intended to support growers, merchants and brewers through their various decision making activities in agribusiness. The first concept results validate the application of a model for future research.

Key words: Applied computer modelling, decision support systems, forecasting, IHGC, information management.

 

Resumen

Diversos agentes de automatización pueden estar involucrados en un moderno sistema de apoyo a la toma de decisiones, con base en los siguientes planteamientos de diseño: Agente E-Learning, Agente de monitoreo, Agente de Diagnóstico de Fallas, Agente de Información del Clima, Agente E-Comercio y Agente de Logística. Los agentes propuestos contribuyen a la educación de los agricultores, a la corrección oportuna de fallas de diagnóstico y a la incorporación de técnicas agrícolas modernas. Además, permiten el procesamiento de información histórica, considerando diversos factores a fin de ayudar en la toma de decisiones. La idea del modelo preliminar "Agentes de Automatización para Sistemas de Apoyo a la Toma de Decisiones" (AASATD) puede aplicarse a varias actividades de agronegocios a nivel internacional. Asimismo, se propone un agente operador que utilice información tecnológica y económica obtenida de todos los Agentes de Automatización mencionados, con el fin de tomar decisiones relativas a la propuesta de posibles escenarios en una cadena de producción que abarca desde la materia prima hasta el producto final. La International Hop Grower's Convention (IHGC) ha lanzado una iniciativa para desarrollar e introducir paulatinamente un modelo de sistema de ayuda a la toma de decisiones para la industria del lúpulo, mediante agentes de automatización incrustados. Está dirigido a apoyar a agricultores, comerciantes y cerveceros en sus diversas actividades de toma de decisiones en la agroindustria. Los resultados del primer concepto validan la aplicación de un modelo para investigaciones futuras.

Palabras clave: modelización computacional aplicada, sistemas de ayuda a la toma de decisiones, IHGC, manejo de la información.

 

INTRODUCTION

Difficulty for decision making in modern agribusiness has increased significantly, since it involves a large number of strongly interrelated factors that affect the satisfaction of the performance criteria describing product quality, production timing and cost. In addition, decision making has to take into account regulations and restrictions concerning the safety of the personnel, the environmental protection and the energy saving. Moreover, agriculture is becoming more commercialized, as farmers are competing with other farmers all over the world. To face these challenges modern agricultural practices must be adopted; however, they require appropriately educated and informed farmers (Abdon and Raab, 2004). So, the question is how to provide the required knowledge and information to farmers, even to those without a high education level.

A powerful tool to circumvent these difficulties is the technological area of agroinformatics and concerns the use of Information Management and Decision Support Systems (IMDSS). They aim to monitoring all functions of an agricultural process and facilitating decision making by proposing scenarios towards satisfying specific performance criteria and restrictions. IMDSS may perform several operations: monitoring the agricultural process, action planning and proposal of scenarios, processing of measurement data to extract information regarding the production cost and the product quality, fault diagnosis and alarm management.

Decision Support Systems (DSS) have been extensively used in industrial applications to support the human-supervisor decisions regarding assurance of efficient and safe processes operation (Lambert et al., 1999; Sanchez et al., 1996). The degree of automation in decision making is the major factor of differentiation between DSS. The DSS characterized by the lower degree of automation simply facilitate decision making by offering information to the operator; in an upper stage DSS incorporate decision-making units that simply propose actions without the jurisdiction of activation. DSS classified in the highest degree of automation completely replace - in certain activities - the human operator. The development of Automation Units for industrial decision making, being implemented as software agents, which may be incorporated in an abundance of commercial Supervisory Control and Data Acquisition (SCADA) products, has been proposed (Koumboulis and Tzamtzi, 2005). The proposed Automation Agents for Decision Support Systems (AADSS) may cover a wide range of industrial applications, providing decision support of the highest degree of automation.

IMDSS specifically oriented for agribusiness applications are met both in the international practice and literature (McCown, 2002; Parrott et al., 2003). Most of them are designed to serve specific sectors of agribusiness, like cotton management (McCown, 2002), vegetable processing industry (Berlo van, 1993), soybean management (Welch et al., 2002), wheat cultivar selection and fertilization (McCown, 2002), irrigation (Mira da Silva et al. , 2001), etc. Moreover, in most cases the proposed DSS are designed to support only a limited range of the decisions to be taken by the farmer. For example, a DSS called PCYield is designed to answer the following two questions aiming to support decisions concerning soybean cultivar selection: What yield range might be expected? And, what happens if irrigation is withheld for a time and the weather is dry? (Welch et al., 2002). An interesting case is DSS supporting farm planning (Recio et al., 2003), which involves a significant range of farming decisions, like scheduling of field tasks, investment analysis, machinery selection and cost/benefit analysis.

Increasing the functionality of DSS with additional characteristics have been proposed, for the example the use of operational research and management's science tools in order to integrate weather forecasts in decision making (Recio et al., 2003). Of special interest is the development of DSS that incorporate tools from agribusiness logistics (Berlo van, 1993; Folinas et al., 2003; Biere, 2001).

Several approaches have been used for developing DSS in agribusiness, based either on operational research/management science, on heuristic search or other artificial intelligence techniques (Recio et al. , 2003). Three of the most popular are linear programming dynamic programming and model-based simulation approaches; a fuzzy logic based approach has also been proposed (Thangavadivelu and Colvin, 1997) for scheduling tillage operations. Despite the research efforts for the development of DSS for agribusiness, farmers seem to be reluctant to involve DSS in their work; thus, the range of application of DSS in agriculture remains significantly smaller than in industry. According to McCown (2002), the two variables long recognized as key to user acceptance is perceived usefulness and ease of use.

The present work proposes stepwise development of Automation Agents for Decision Support Systems (AADSS) for agribusiness. Following Koumboulis and Tzamtzi (2005) for industrial applications, the automation agents will be implemented as software units, which may be incorporated in several commercial SCADA products. The proposed agents aim to support decision making in a significant range of the agribusiness operation, extending from cultivation techniques to farm planning and commerce of products, by undertaking to execute actions, like monitoring the agricultural process, providing e-learning functionalities, fault diagnosis, e-commerce support, planning based on logistics and processing of weather information. Thus, the farmer is strongly supported with regard to all decision making that concerns the actions to be performed for all stages of agribusiness, from production to commerce. Moreover, the AADSS will provide a very friendly and easy to use graphical user interface, exploiting the graphical user interface capabilities of commercial SCADA products.

 

MATERIALS AND METHODS

Automation agents

The proposed Automation Agents will be based on modern techniques of sustainable agriculture, so as to face the restrictions and difficulties of agricultural environment, such as high complexity, presence of uncertain and time changing factors, like weather, and restriction of natural sources, like water.

The proposed Automation Agents are the Supervisory Agents and the Operator Agent. The Supervisory Agents are the E-learning Agent, the Monitoring Agent, the Fault Diagnosis Agent, the Weather Information Agent, the E-Commerce Agent and the Logistics Agent. Their functionality aims to supervise all aspects of agribusiness, from production to commerce, and moreover to provide e-learning services to farmers' group. It is important to note that the knowledge data base used for the E-learning Agent is dynamically adapted with any new information derived from processing the data gathered from the process. The Operator Agent exploits information gathered from all the aforementioned Automation Agents in order to support the farmer's decision making by proposing possible scenarios to the farmer. The interconnection between the Operator and the Supervisory Agents is illustrated in Figure 1.

Operator Agent

The Operator Agent will be implemented according to the DAI-DEPUR architecture, an integrated and distributed artificial intelligence supervisory architecture, proposed by Sanchez et al. (1996) for a waste-water treatment plant. The DAI-DEPUR architecture has also been used for the implementation of an Operator Agent for industrial applications (Koumboulis and Tzamtzi, 2005). Below we present in short the DAI-DEPUR architecture as described by Sanchez et al. (1996).

The DAI-DEPUR architecture comprises the data level, the distributed knowledge level, the reasoning level and the supervisory level.

The data level comprises the data collection system, that receives data from the process sensors, and the data base management system. The distributed knowledge level comprises distributed agents each of which processes validates and monitors the available information for a specific subsystem of the process, in order to describe the subsystem's behavior. Their conclusions are sent to the supervisory level in order to diagnose the whole plant state. The distributed knowledge level comprises simulation modules, as well as knowledge acquisition modules. The role of the knowledge acquisition modules is two-fold; first a conceptual clustering of data is performed, that leads to a representation of the process domain of operation in terms of classes, and then conjunctive classification rules are determined.

The reasoning level manages a Case Library that contains information about previously experienced situations, as well as solutions that have been taken in the past for each of these situations. Besides, the reasoning level may evaluate proposed solutions using simulation. The Case Library is dynamically enriched with information regarding the newly experienced situations.

The supervisory level gathers information from the distributed knowledge and the reasoning level, to determine the current status of process operation and coordinate the rest of Automation Agents. In our case the supervisory level is responsible for all decisions undertaken by the Operator Agent, that is:

1) Alarm management

2) Exchange of information between the Automation Agents

3) Configuration of the supervisory agents

4) Derivation and proposal of optimal, or at least suboptimal scenarios of actions to be undertaken by the farmer, based on specific performance requirements.

Due to the complexity of the agricultural processes, in conjunction with the large number of factors to be taken into account, it is a usual case in agribusiness to deal with competitive goals that is criteria that cannot be simultaneously achieved. For example, the improvement of the product quality usually increases the production cost; then scenarios proposed to the farmer should be based on a compromise between competitive design goals, performed to achieve the optimal result, regarding the economic issues of the process, like

production cost and product quality, as well as energy or natural resources saving, while satisfying constraints imposed by environmental protection rules. This compromise may be formulated as an optimization under constraints problem (Koumboulis and Tzamtzi, 2005).

Supervisory Agents

Six Supervisory Agents are proposed based on the following approaches, selected to serve the needs of agribusiness:

1) E-Learning Agent

2) Monitoring Agent

3) Fault Diagnosis Agent

4) Weather Information Agent

5) E-commerce Agent

6) Logistics Agent

The E-Learning Agent aims to educate farmers with regard to required information and knowledge concerning the production techniques and the commerce of the specific product. The E-Learning Agent comprises a historic module implemented by a database, and a knowledge processing unit. The historic module contains required information background in order to make decisions concerning the agricultural process. The historic module stores measurements regarding the agricultural process and the weather, technical and scientific information concerning for example cultivation methods, plant diseases, special characteristics of each plant variety, effect of the weather on the plant, etc.; market information, like the available stock for each plant variety, current prices, current request from a national or a global market, etc., as well as legislation rules and restrictions. The knowledge processing unit is used in order to dynamically adapt the historic module based on the newly gathered information. The E-Learning Agent exploits also the information stored in the Case Library of the Operator Agent's reasoning level concerning previously experienced situations. The Monitoring Agent aims to provide the farmer, as well as the other Agents, all the necessary information regarding the current status of the agricultural process. For this purpose, the Monitoring Agent exploits the data level of the Operator Agent that collects information from the agricultural process sensors, like measurements of humidity and temperature, height of the plant, etc. These data are processed in order to derive information about the current status of the agricultural process, as for example the plant growth. The information provided by the Monitoring Agent is also used by the distributed knowledge level of the Operator Agent.

The Fault Diagnosis Agent aims to a timely diagnosis of any fault that may occur in the agricultural process, like plant diseases, harm caused by insects, malfunctioning of the irrigation system, problems due to weather conditions, etc. Fault detection is achieved using the information provided by the Monitoring Agent and it is based on specific fault detection rules provided by the historic module of the E-Learning Agent. Whenever a specific situation deviates from the rules provided by the historic module of the E-Learning Agent, data are either processed by the knowledge process unit of the E-Learning Agent or they are sent to an expert, in order to derive new rules that will enrich the fault diagnosis capabilities of the Agent. The Fault Diagnosis Agent may also use the functionality of the distributed knowledge level of the Operator Agent.

The Weather Information Agent aims to support the decision making process regarding the issues that concern weather conditions. This Agent gathers information from meteorological services, combining it with measurements from the cultivation site. Based on this information it provides short term prediction of weather conditions. Moreover, the cooperates with the historic module of the E-Learning Agent in a two fold manner: first it exploits information from the historic module to support the weather prediction process; secondly, the information gathered by him is used to enrich the historic module.

The E-Commerce Agent aims to support the sales through internet of agricultural products and it can also support the farmer in buying row material through internet. The E-Commerce Agents comprises an intelligent unit to use the market information stored in the historic module, in order to decide whether each buying or selling action is profitable for the farmer.

The Logistics Agent aims to support the farmer concerning decisions on planning and controlling an efficient flow and storage of raw materials, intermediate and final products from point of origin to point of consumption, so as to achieve specific performance requirements, while simultaneously satisfying constraints (Berlo van, 1993; Folinas et al., 2003; Biere, 2001). The performance requirements concern the product cost and quality, the satisfaction of the market demand, the optimal exploitation of the processing equipment and cultivation area, etc. The restrictions concern the storage capacity, the maximum allowed time of storage, the available cultivation capacity and the processing capacity, etc. The planning provided by the Logistics Agent should take into account time changing and uncertain factors, like the weather conditions and the demand of the market. The Logistics Agent may consider the whole logistical chain incorporating agriculture, processing industry and market, as was proposed by Berlo van (1993); in this case, the DSS should communicate with corresponding information systems of the processing industry and market sectors, in order to derive the data required.

Embedding automation agents in SCADA systems

The implementation of Automation Agents should satisfy certain functional specifications, like interactive communication, real time processing and precision of calculations. To achieve these goals, the software development of the Automation Agents is based, following Koumboulis and Tzamtzi (2005), on international standards for open architecture, which will assure the embedment of the software in several commercial SCADA products. These systems are modern Information Systems with build-in capabilities of network communication that have widely been used in modern industry. The subsystems of a SCADA system are:

1) Bilateral communication with the process: data acquisition from sensors and command transmission though appropriate actuators,

2) Graphical User Interface (GUI), through which the operator monitors and commands the process.

3) Automation Agents for Decision Support System.

The AADSS functions, provided by most of the commercial SCADA products, are: storage of info to historical modules, statistical data processing, formulation of reports, and alarm management. The Automation Agents presented in the previous sections will utilize the network communication, collection, registration, depiction and data process capabilities provided by modern SCADA. For example, the data level of the DAI-DEPUR architecture will be implemented by the data processing unit of the SCADA system, while the alarm management operation performed by the supervisory level of the DAI-DEPUR architecture will be supported by the alarm handling unit of the SCADA system. Besides communication of Automation Agents with the human-operator will take place via the graphical user interface of SCADA system. The embedment of the Automation Agents in SCADA systems is achieved using the latest technology of modern software tools, supported by most of modern commercial SCADA programs, like the OPC (OLE for Process Control) that allows bilateral real time data transfer between different types of equipments, as well as established standards of object-oriented communication between heterogeneous applications of Windows operating systems, like COM and ActiveX. The embedment of Automation Agents into SCADA systems is presented (Figure 1).

 

RESULTS AND DISCUSION

Hop industry case study

An implementation of the proposed scheme will take place within a framework of a running research project for a case study in a hop industry.

Hops are one of the four main ingredients (malt, yeast, water) in the brewing industry which influence directly a hop demand. Main functions of hops are to give beer a distinctive aroma and to provide bitterness to a final beer. Adding different hop varieties to the wort kettle produces the typical beer aromas.

The hop industry must establish its informative and useful information system due to the following reasons: 1) hops are a classic internationally agricultural traded commodity, one of the few internationally traded goods bought and sold on world markets without any major economic restrictions, that is on the real basis of supply and demand; 2) the demands on production techniques, varieties, quality and preparation of export producing hops is changing constantly (Pavlovic et al. , 2003; Pavlovic and Koumboulis, 2004). For the sake of international character of hops as well as stakeholders' initiatives, the International Hop Growers' Convention (IHGC) with 27 companies and organisations from 19 countries took over an initiative in 2006 to improve the information management for a benefit of its members (Figure 2).

Furthermore, there will also be benefits for the brewing industry. The seven largest breweries (InBev, 12.6%; SABMiller, 11.0%; Anheuser-Busch, 10.9%; Heineken, 7.4%; Carlsberg, 3.0%; Molson-Coors, 3.0%; Modelo, 2.9%) produced more than 50% of the world beer in 2005, but there are many small scale breweries that may benefit from the AADSS model information results. Regardless the size of a brewing company, the product principles as well as the objectives regarding technology and purchasing are more or less the same: to use raw materials as economically as possible in the brewing process (Pavlovic et al., 2006).

 

CONCLUSIONS

The proposed Automation Agents are generalised tools applicable to any agribusiness, but the implementation of these Agents should be strongly directed to specific plants or families of plants. The proposed AADSS is particularly suited for large agricultural production units, since for small units the decision making process is significantly easier. Of special interest are farms with several plots of different crops, where the global farm resources have to be appropriately distributed among the plots, as well as farms simultaneously coordinated, as is the case of agricultural cooperatives.

The proposed Agents may contribute to farmers' education, to early correct fault diagnosis and incorporation of modern agricultural techniques, to process historical data taking into account several factors in order to support decision making, and to support the farmer in e-commerce activities. The incorporation of Automation Agents in DSS will contribute considerably in the decongestion of the responsibilities of the farmer. Thus farmers are assisted to focus on higher level operations like production planning and commerce. Besides, production costs are decreased, as the performance criteria can be achieved in a more precise manner and with less human effort. Finally, the environmental protection may benefit significantly since the proposed agents will contribute to a better exploitation of natural sources, as well as energy saving. The implementation of the Automation Agents is based on open architectures, so as to give the capability of integration in a crowd of commercial SCADA systems, fact that simplifies significantly the development and implementation procedure of the Automation Agents. What is more, the graphical user interface of SCADA systems provides a user friendly environment that will increase the farmers' acceptance.

 

ACKNOWLEDGEMENT

This paper has been funded by the running Eureka Project E! 3219 - AADSS, EU.

 

LITERATURE CITED

Abdon, B. R., and R. T. Raab. 2004. Knowledge sharing and distance learning for sustainable agriculture in the Asia Pacific Region: The role of Internet. New directions in a diverse planet. Proc. of the 4th Int. Crop Science Congress, Brisbane, Australia, Sept. 26. - 10. Oct. 2004. http://www.cropscience.org.au/icsc2004/        [ Links ]

Berlo van, J.M. 1993. A decision support tool for the vegetable processing industry; an integrative approach of market, industry and agriculture'. Agric. Systems 23: 91-109.         [ Links ]

Biere, A.W. 2001. Agribusiness logistics: An emerging field in agribusiness education'. In: IAMA World Food and Agribusiness Symposium. Sydney, Australia, 27. June, 2001. http://www.ifama.org/conferences/2001Conference/Papers/Area%20I/Biere_Arlo.PDF        [ Links ]

Folinas, D., M. Vlachopoulou, V. Manthou, and B. Manos. 2003. A web-based integration of data and processes in the agribusiness supply chain. EFITA 2003 Conference. Debrecen, Hungary: 143-149.         [ Links ]

Koumboulis, F. N., and M. P. Tzamtzi. 2005. Automation Agents Embedded in Industrial Decision Support Systems. In: International Conference on Intelligent Agents, Web Technologies and Internet Commerce - IAWTIC'2005. Vienna, Austria. pp: 51-57.         [ Links ]

Lambert, M., B. Riera, and G. Martel. 1999. Application of functional analysis techniques to supervisory systems. Reliability Eng. & System Safety 64: 209-224.         [ Links ]

McCown, R. L. 2002. Changing systems for supporting farmers decisions: Problems, paradigms, and prospects. Agric. Systems 74: 179-220.         [ Links ]

Mira da Silva, L., J. R.Park, J. D. H. Keatinge, and P. A. Pinto. 2001. A decision support system to improve planning and management in large irrigation schemes. Agric. Water Manag. 51: 187-201.         [ Links ]

Parrott, L., R. Lacroix, and K. M. Wade. 2003. Design considerations for the implementation of multi-agent systems in the dairy industry. Computers and Electronics in Agric. 38: 79-98.         [ Links ]

Pavlovic, M., F. N. Koumboulis, M. P. Tzamtzi, and D. A. Karras. 2003. Model of information management and data exchange on a global hop supply. Proc. of the 2^nd Int. Symp. on Intelligent Information Technology in Agriculture. Beijing, China, pp. 20-23.         [ Links ]

Pavlovic, M., and F. M. Koumboulis. 2004. Methodology of an IHGC Market Supply Data Collation. Hop Bull. 11: 17-24.         [ Links ]

Pavlovic, M., X. Luo, I. J. Kosir, M. Virant, and F. Gu. 2006. Expansion of the Chinese hop and brewing industry. Hop Bull. 13: 61-69.         [ Links ]

Recio, B., F. Rubio, and J. A. Criado. 2003. A decision support system for farm planning using AgriSupport II. Decision Support Systems 36: 189-203.         [ Links ]

Sanchez, M., U. Cortes, J. Lafuente, I. R. Roda, and M. Poch. 1996. DAIDEPUR: An integrated and distributed architecture for waste-water treatments plants. Artificial Intelligence in Eng. 1: 275-285.         [ Links ]

Thangavadivelu, S., and T. S. Colvin. 1997. Fuzzy-Logic-Based decision support system for scheduling tillage operations. Eng. Applications of Artificial Intelligence 10: 463-472.         [ Links ]

Welch, S. M., J. W. Jones, M. W. Brennan, G. Reeder, and B. M. Jacobson. 2002. PCYield: Model-based decision support for soybean production. Agric. Systems 74: 79-98.         [ Links ]