Progress 08/16/19 to 07/31/22
Outputs
Target Audience:Undergraduate Students: Eight (8) undergraduate students (Computer science major) participated in this project. The student researchers participated in implementing the proposed algorithms in bash script, setting up simulations, conducting experiments, and producing performance graphs. They learned about different cyber-attacks in IoT and the importance of securing a smart farming system. The student researchers also developed strong programming skills and honed their skills and knowledge on cybersecurity topics including intrusion detection, port scanning, and moving target defense. Moreover, they developed various professional skills including communications, and the ability to work effectively in a team. This project provided an inclusive and collaborative research environment by recruiting students from different backgrounds. Two international students participated as undergraduate researchers in this project. Moreover, it promoted the participation of women in science and technology by providing research experience to four (4) female undergraduate students. IoT/Cybersecurity/Agricultural Researchers: The optimization models, algorithms, and experimental results generated during this project were presented through technical reports, posters, and conferences. As cyber security in agriculture is still an under-explored topic, our research findings drew the attention of the audience and led to productive research discussions and identification of potential agricultural use-cases. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?This project provided eight (8) undergraduate computer science students with the opportunity to engage in various research activities of this project. The PI provided training on python, Linux Bash Scripts, and cybersecurity tools including Kali Linux, Snort, hydra, aircrack through multiple hands-on sessions. The PI prepared tutorials on these topics and shared them with the students. The tutorials provide step-by-step instructions on using security tools to learn different cyber-attacks such as password attack, and de-authentication attacks. The students also learned about using defense tools such as snort to detect intrusions. This project also allowed the students to develop their professional skills as the students were required to give presentations in weekly research meetings. The students were required to present progress on their tasks including pseudo-codes, programs, and performance graphs. The students also received training from the PI on creating performance charts and data analysis in Excel. We provided a co-authorship opportunity to our student: "Kristin Barrett" in our conference article, where she contributed to developing java code in ifogsim simulator for comparing different schemes to place the IoT applications. This project provided an inclusive and collaborative research environment that allowed students to exchange their ideas and develop the ability to work effectively in a team. We also provided a technical presentation on IoT to an 1890 extension agent at South Carolina State University to increase awareness of smart farming technology. How have the results been disseminated to communities of interest?We have disseminated our research products and findings to students, faculty members from SCSU and other institutions, industry professionals, and researchers through our presentation in technical meetings, conferences, and student research events. We also stored our articles in an open-access digital archive for faster dissemination. The dissemination methods used in this project are described as follows: 1) Technical Meetings: The PI has presented the research results in the following meetings that included faculty members from SCSU and professionals/researchers from Savannah River National Laboratory (SRNL). a) Agriculture Cyber/robotics/AI/ML using Big Data Meeting (August 21, 2021): The research work, especially the Deep Reinforcement Learning (DRL) algorithm for designing a robust defense mechanism for smart farming drew the attention of SRNL researchers. The PI engaged with the SRNL researchers in a discussion about the implementation of the DRL algorithm, attacker's model, and experimental methods. b) Cyber/Computer Science Collaboration SCSU/SRNL Meeting (June 25, 2021): The PI presented IoT technology and the benefits of an IoT-based smart farming system. The PI also engaged in discussions with the SRNL team about the potential applications of smart farming and the pertinent issues for its deployment in South Carolina. 2) Conferences: Our research results have been disseminated to IoT/cybersecurity/agricultural science research community through our publication in conferences (IEEE Southeastcon 2021, MCSMS 2021, EAI Edge-IoT 2021), presentation in symposiums (1890 ARD Symposium, 2022), and poster presentation (WiCyS 2022). The conference articles present an IoT application model for smart farming, the ILP model of secure placement of IoT applications, and the fog node placement, and K-means clustering based algorithm. Our presentations at the 1890 ARD Symposium includes MDP model, DRL algorithm, and network diversity-based MTD. One of our students presented a poster on intrusion prediction aware MTD in WiCyS which is a premier cybersecurity conference attended by industry, federal agencies, and academic institutions. The fact that our work on MTD received the best research poster award in WiCyS shows the positive impact this research has made in the cybersecurity research community. 3) Student Events: Our student researcher "Kristin Barrett" presented her research on MTD during a student event organized for the visit of the DOE Secretary Honorable Jennifer M. Granholm and Honorable congressman James Clyburn on Feb. 17, 2022 at South Carolina State University. Kristin was interviewed by the press on her interest and work in the cybersecurity field. We also shared our research results with students and faculty members from other universities such as the University of North Carolina Charlotte (UNCC). Kristin presented her research during the Intelligence Community- Center for Academic Excellence (IC-CAE) student researcher event held on April 22, 2021. The event was organized by the UNCC and was attended by students and faculty members from SCSU and UNCC. 4) Open Access Digital Archives: To ensure early dissemination of our research findings, we uploaded our articles on IoT application model, and IoT application placement into arXiv, an open access digital archive maintained by Cornell University. What do you plan to do during the next reporting period to accomplish the goals?
Nothing Reported
Impacts
What was accomplished under these goals?
The research activities of this project involved the design of new mathematical models, efficient algorithms, experimentation, and developing a prototype of a smart farming system. The details of the activities, experiments, and key observations are described as follows: Goal (1): We developed mathematical models of the IoT gateway placement (also called fog node placement) problem and IoT application placement problem using ILP. Our objective of gateway placement was to optimize the cost of deploying IoT gateways (also referred to as fog nodes) in a farm area to minimize the energy cost of sensors and ensure the sensor workload is evenly distributed among the fog nodes. In application placement, we aim at minimizing the resource cost and ensuring the quality of service (QoS) requirements such as security and latency are met by the placement. Goal (2): We designed novel and efficient fog node placement algorithms using K-means clustering and Hedonic games. These algorithms can group the sensors deployed in a farm area into K clusters in a way that reduces the energy consumption and ensures effective resource utilization. We also designed an IoT application placement algorithm using genetic algorithm (GA) to find a cost-effective mapping of smart farming applications with appropriate resource nodes (e.g., cloud server, fog server) while providing a guarantee of the satisfaction of QoS and security requirements of the smart farming applications. Goal (3): To understand the effectiveness of the moving target defense (MTD) approach, we first investigated proactive and reactive MTD schemes to protect IoT nodes from cyber-attacks. We focused on IP shuffling, port shuffling and MAC address shuffling and observed their effectiveness in preventing SSH password attacks and Wi-Fi de-authentication attacks. The above works provided us with enough insights to design Markov Decision Process (MDP) models to design MTD schemes that focus on IP shuffling and safe migration of VMs among the fog nodes to obfuscate their identity from attackers. We modeled the states, actions, and the reward functions of our MDP models by considering a threat scenario for smart farming. Goal (4): We designed Deep Reinforcement Learning (DRL) algorithms using the Deep Q Network (DQN) to solve our MDPs. DQN is based on two key methods: experience replay and network cloning. In the experience replay method, the agent stores its experience of transitions in an experience memory. When training the neural network, the transitions are sampled from the memory, allowing the agent to use its past experience in updating the Q-value. In the network cloning method, the agent uses two identical networks: a target network and an evaluating network. DQN takes states as the input and provides the optimal security action as the output. Goal (5): We conducted experiments to evaluate the performance of the proposed placement algorithms. We evaluated the K-means based fog node placement (k-FNP) algorithm using three performance metrics: 1) Energy Consumption, 2) Standard Deviation of Load, and 3) Maximum Load. Our simulation results show that the k-FNP algorithm significantly outperforms the random algorithm in terms of all three metrics. One notable observation is that placement obtained using K-FNP algorithm consumes 13 times less energy than a random placement algorithm. Another interesting observation shows 69% better performance by k-FNP compared to the random placement. We also evaluated the convergence performance of GA-based application placement. Our results show that the algorithm can find the best solution in just a few iterations, which ensures the computational efficiency of the algorithm. Goal (6): We evaluated the proposed DRL algorithms using their reward performance. Our experiment consisted of 100 episodes of training the DRL agent. We measured the reward function for different learning rates. We observed that reward increases fast for learning rates: 0.02 and 0.05. The agent achieves convergence at around the 20th episode for both learning rates. However, the learning rate of 0.02 results in a better reward value than the learning rate of 0.05. The early convergence behavior of our DRL agents confirms the ability of the algorithm to determine optimal security action (e.g., VM migration schedule) in real-time. Goal (7) We built a prototype of IoT-based smart farming system, called farm monitor that is made up of Raspberry PI devices, sensors, edge nodes, and a cloud server. The farm monitor provides temperature, humidity, soil moisture, and soil temperature values in real-time. We used python to develop the IoT applications. Our prototype makes use of edge nodes for reducing the communication latency and a cloud server for scalability. We also designed a dashboard using Node-Red service to display the sensor readings using gauge and line charts. Goal (8) We deployed the prototype in our lab and tested its features using a wireless client that connects to the same wireless network as the edge node and connects to the cloud server over the Internet. Our evaluation shows that the client successfully receives the sensor readings from the edge node. We also observe that the client can obtain current as well as historical observations on the cloud server that will enable the client (e.g., farmer) in making critical agricultural decisions efficiently.
Publications
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2021
Citation:
1) J. Sahoo, Optimal Secure Placement of IoT Applications for Smart Farming , Proceedings of The 7th International Workshop on Mobile Cloud Computing systems, Management, and Security (MCSMS2021), Dec. 6- Dec. 9, 2021, Gandia, Spain.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2021
Citation:
2) J. Sahoo, Energy and Load Aware Fog Node Placement for Smart Farming , Proceedings of EAI Edge-IoT 2021-2nd EAI International Conference on Intelligent Edge Processing in the IoT Era, Nov. 24 26, Portugal.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
4)J. Sahoo, M. Islam, Robust and Dynamic Scheduling of Smart Farming Applications using Deep Reinforcement Learning , 1890 ARD Research Symposium, Atlanta, GA, April 2-6, 2022.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
5) K. Barrett, T. Clark, J. Sahoo, Network Diversification for Preventing Cyber-Attacks in Smart Farming: An Experimental Evaluation , 1890 ARD Research Symposium, Atlanta, GA, April 2-6, 2022.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
6) J. Sahoo, Optimal Secure Placement of IoT Applications for Smart Farming , Presented at the 7th International Workshop on Mobile Cloud Computing systems, Management, and Security (MCSMS 2021), Dec. 6- Dec. 9, 2021, Gandia, Spain (Online).
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2021
Citation:
3) J. Sahoo and K. Barrett, Internet of Things (IoT) Application Model for Smart Farming, Proceedings of IEEE SoutheastCon 2021, pp. 1-2, doi: 10.1109/SoutheastCon45413.2021.9401845.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
7) J. Sahoo, Energy and Load Aware Fog Node Placement for Smart Farming , Presented in EAI Edge-IoT 2021-2nd EAI International Conference on Intelligent Edge Processing in the IoT Era, Nov. 24 26, Portugal (Online).
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
8) K. Barrett, J. Sahoo, Moving Target Defense for Securing Internet of Things , presented at the Intelligence Community- Center for Academic Excellence (IC-CAE) student researcher event, April 22, 2021, (Virtual).
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
9) J. Sahoo and K. Barrett, "Internet of Things (IoT) Application Model for Smart Farming," Presented at the IEEE SoutheastCon, March 2021. (Online)
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
T. Clark, K. Barrett, J. Sahoo, Intrusion Prediction Aware Moving Target Defense for Smart Farming , 2022 Women in Cybersecurity Conference, March 17-19, Cleveland, Ohio. (Received the Best Research Poster Award)
|
Progress 10/01/20 to 09/30/21
Outputs
Target Audience:Three undergraduate students (Computer science major) participated in this project. The student researchers participated in implementing the proposed algorithms in bash script, setting up simulations, conducting experiments, and producing performance graphs. They learned about different security issues in IoT and the importance of securing a smart farming system. The student researchers also developed strong programming skills and honed their skills and knowledge on cybersecurity topics including intrusion detection, port scanning, and moving target defense. Moreover, they developed various professional skills including communications, and the ability to work effectively in a team. This project provided an inclusive and collaborative research environment by recruiting students from different backgrounds. Moreover, it promotes the participation of women in science and technology by providing research experience to three female undergraduate students. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?-This project provided three (3) undergraduate computer science students with the opportunity to engage in various research activities of this project. The PI trained the students on writing Linux Bash Scripts, and using cybersecurity tools: Kali Linux and Snort through multiple demo sessions. The PI prepared tutorials on these topics and shared them with the students. The tutorials are designed in a way to help students develop the skills to use the tools in practical security scenarios. -This project also allowed students to develop their professional skills through their research presentations in weekly research meetings. The students were required to present progress on their tasks that included algorithms, programs, and performance graphs. The students also received training from the PI on creating charts and data analysis in Excel. -Our student: "Kristin Barrett" had the opportunity to be a co-author in a conference article. She also presented her research in two research venues: 1) IEEE Southeastcon 2021, and 2) Student researcher event. Southeastcon is a reputed IEEE Conference and Kristin's presentation was well-appreciated by the conference attendee. The student researcher event was hosted by Intelligence Community- Center of Academic Excellence, University of North Carolina, Charlotte. Kristin's presentation on moving target defense drew attention of the attendees as some of them showed interest in knowing more about our research work. How have the results been disseminated to communities of interest?1)The PI has presented the research results in the following meetings that included faculty members from SCSU and professionals/researchers from Savannah River National Laboratory (SRNL). a)"Agriculture Cyber/robotics/AI/ML using Big Data Meeting" (August 21, 2021): The research work, especially the Deep Reinforcement Learning (DRL) algorithm for designing a robust defense mechanism for smart farming caught the attention of SRNL researchers. The PI engaged with the SRNL researchers in Q/A on the implementation of DRL algorithm, attacker's model, and experimental methods. b) "Cyber/Computer Science Collaboration SCSU/SRNL Meeting" (June 25, 2021): The PI discussed about IoT technology and the benefits of a smart farming system made up of IoT devices. The PI also engaged in discussions with SRNL team about the potential applications of smart farming and the pertinent issues for its deployment in South Carolina. 2) We have disseminated our research results to IoT research community through our publication in three conferences. The conference articles present IoT application model for smart farming, ILP model of secure placement of IoT applications, ILP model of fog node placement, and K-means clustering based algorithm. What do you plan to do during the next reporting period to accomplish the goals?We plan to conduct the following activities in the next reporting period to accomplish goals (7), (8), and (9): 1) We will continue conducting security experiments to study the efficiency of moving target defense for a smart farming system. So far, we have measured the convergence performance of the DRL algorithm. Our next experiments will focus on measuring the security performance i.e., the ability of the algorithm to prevent cyber-attacks. 2) We will build a prototype to measure the performance of the smart farming system in a real setting. The prototype will involve a cloud and fog infrastructure, IoT gateways, IoT nodes, and an end-user device. Various smart farming applications such as soil management, yield prediction will be implemented and deployed on the cloud-fog infrastructure. These applications provide farmers with the real-time status of the farm, critical alerts, and useful insights to make informed farming decisions. We will evaluate the prototype using performance metrics such as response time, cost, and resource utilization. 3) We have started identifying farmers in the Orangeburg area. Our next step would be to reach out to them and exercise online questionnaire to receive data on types of crops, and challenges they face in their day-to-day agricultural operations. We will host a workshop for the farmers to make them aware of the IoT technology and the cybersecurity threats in smart farming. ?
Impacts
What was accomplished under these goals?
Our research activities involved design of new mathematical models, algorithms, and simulation experiments that address the placement and security need of a smart farming system. We developed a security model using Markov Decision Process (MDP) by incorporating the security requirements of IoT nodes. The model can serve as a basis to develop advanced security models for complex smart farming scenarios involving heterogeneous sensors. We also designed a robust defense mechanism that incorporates a learning agent that can learn to optimize its strategies by observing the attacker's behaviors and can make better security decisions. The proposed defense mechanism can prevent cyber-attacks such as man-in-the-middle attacks, denial of service, and malware. As a result, farmers can leverage the benefits of smart farming including improved yield, enhanced decision support, and reduced cost without any fear of data breaches or hacking of their IoT devices. This project provided research experiences to three undergraduate computer science students and helped them enhance their creativity and analytical skills. Moreover, the student researchers enhanced their programming skills by being engaged in extensive simulations that required them to write multiple Linux bash scripts. Also, we have published our research results in three conference papers. We have presented our work at the IEEE Southeastcon 2021 conference which is a reputed regional conference. We have also presented our research in two international conferences: EAI Edge-IoT 2021 and IEEE IoTSMS 2021. In addition, we have produced one student poster (accepted in WiCyS 2022 conference) on our work on Smart Farming security. Goals Accomplished: This year, we accomplished the goals (3), (4), (5) and (6). The activities, experiments, and key outcomes are described as follows: Goal (3): To understand the effectiveness of MTD approach in protecting IoT nodes from cyber-attacks, we conducted experiments in a virtual environment. We considered three types of MTD: IP shuffling, port shuffling and MAC address shuffling and implemented these schemes using linux bash script. We also implemented the attacker's probing strategy in bash script. The experiments provided us with insights about the working of MTD and also the importance of Intrusion Detection System (IDS) in detecting intrusions and triggering MTD when needed. We also designed a proactive and reactive MTD schemes using time-series analysis. We modeled the intrusions as a time-series and forecast them using Exponential Moving Average (EMA) method. We evaluated the performance of proactive and reactive MTD schemes. We considered "attack success rate" as the performance metric that denotes the number of times the attacker discovers the ports of the critical services (e.g., File Transfer Protocol (FTP)). We observed that the proposed proactive MTD performed better than a basic proactive MTD that changes the configuration using either a fixed or random time interval. Similarly, we observed that the proposed reactive MTD outperformed a basic reactive MTD, thereby offering the attacker the scope to attack the system before the MTD is triggered. The above works provided us with enough insights to design different components of Markov Decision Process (MDP). We considered IP shuffling based MTD as it will help prevent the smart farming sensors and gateways from being discovered by the attacker and avoid serious attacks such as Denial of Service and service Brute-force attacks. We modelled the states, actions and the reward function for a smart farming scenario. The state represents the status (safe, unsafe) of an IoT node at a given time. The action in our MDP represents the difference between old IP address and the new IP address. Our intuition is that by optimizing the difference, it will be difficult for the attacker to discover the IoT nodes, thereby preventing cyber-attacks. Goal (4): We designed a DRL algorithm using the Deep Q network (DQN) to solve our MDP. DQN is based on two key methods: experience replay and network cloning. In the experience replay method, the agent stores its experience of transitions in an experience memory. When training the neural network, the transitions are sampled from the memory, allowing the agent to use its past experience in updating the Q-value. In the network cloning method, the agent uses two identical networks: a target network and an evaluating network. DQN takes states as the input and provide the optimal action (i.e., the gap between old IP address and new IP address) as the output. Goal (5): We conducted simulation experiments to evaluate the performance of K-means based fog node placement (k-FNP) algorithm. The objective of the algorithm is to find the location of fog nodes that minimizes the cost i.e., energy consumption of IoT nodes, ensure fairness in allocating sensor workload among the fog nodes. We compared the proposed algorithm with a random placement algorithm using three performance metrics: 1) Energy Consumption, 2) Standard Deviation of Load, and 3) Maximum Load. Our simulation results show that our algorithm significantly outperform the random algorithm in terms of all three metrics. The highest value of energy consumption in random algorithm is 12X109 joules, whereas the proposed algorithm k-FNP only requires 9X108 joules. In terms of standard deviation of load, the random algorithm shows a 360% increase over k-FNP, and hence it fails to ensure efficient utilization of fog resources. k-FNP yields 69% better performance than random placement in maximizing the load for a small-scale scenario, whereas for a large-scale scenario, random placement shows 4.4 times higher load than k-FNP. Goal (6): We implemented the proposed DRL algorithm in Python. By utilizing the neural network, DQN is expected to converge to the optimal policy quickly when there is a large state-action space. We carry out 500 episodes of experiments under direct attacker using different learning rate and temporal discount factor. We observe that with small learning rate, the local minima can be obtained more frequently. We also observe that reward increases very fast and converges to the max reward. This is because DQN avoids critical nodes/devices proactively by learning the attack strategy, which can reduce the attack success rate and ensure resiliency of the smart farming system.
Publications
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
1) J. Sahoo, Optimal Secure Placement of IoT Applications for Smart Farming, Presented in The 7th International Workshop on Mobile Cloud Computing systems, Management, and Security (MCSMS 2021), Dec. 6- Dec. 9, 2021, Gandia, Spain (Online).
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
2) J. Sahoo, Energy and Load Aware Fog Node Placement for Smart Farming, Presented in EAI Edge-IoT 2021-2nd EAI International Conference on Intelligent Edge Processing in the IoT Era, Nov. 24 26, Portugal (Online).
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
3) K. Barrett, J. Sahoo, Moving Target Defense for Securing Internet of Things , Intelligence Community- Center for Academic Excellence (IC-CAE) student researcher event, April 22, 2021, (Virtual).
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
4) J. Sahoo and K. Barrett, "Internet of Things (IoT) Application Model for Smart Farming," Presented in IEEE SoutheastCon, March 2021. (Online)
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
5) T. Clark, K. Barrett and J. Sahoo, Intrusion Prediction Aware Moving Target Defense for Smart Farming , Accepted for presentation in Women in Cybersecurity (WiCyS) 2022 conference, March 17-19, Cleveland.
|
Progress 10/01/19 to 09/30/20
Outputs
Target Audience:Five undergraduate students (Computer science major) including two female students participated in this project. The student researchers participated in surveying the literatures, implementing algorithms in python/java, setting up simulations, conducting experiments, and producing performance graphs. They learned the important research issues, and research directions in advanced technologies such as IoT, Cloud, and fog computing. The student researchers also developed the strong analytical skills and the ability to design innovative algorithms. Moreover, they developed various professional skills including communications, and the ability to work effectively in a team. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?This project provided five (5) undergraduate computer science student researchers including two female students with the opportunity to engage in various research activities. The PI provided training on the iFogSim simulator, and modeling of IoT applications. This project also allowed students to develop professional skills through their research presentations in weekly research meetings. Overall, the research and implementation experience will help these students in pursuing advanced degrees in computer science. This project provided an inclusive and collaborative research environment by recruiting students from different backgrounds. Moreover, it promotes the participation of women in science and technology by providing research experience to two female undergraduate students. How have the results been disseminated to communities of interest?We have uploaded a technical report entitled "Internet of Things (IoT) Application Model for Smart Farming" to the Cornell university digital library (Link: https://arxiv.org/ftp/arxiv/papers/2101/2101.03722.pdf) for a faster dissemination of our experimental results with the IoT research community. We also integrated the smart farming concepts as well as our research on IoT security in two modules: "IoT/Cyber-Physical Systems" and "Security of IoT/Cyber-Physical Systems" of the "CS 225: Introduction to Cybersecurity", an undergraduate-level course in BS in Cybersecurity concentration at South Carolina State University. Students who took this course developed an understanding of smart farming, security attacks (e.g., man-in-the middle, eavesdropping, etc.) and the techniques to secure the agricultural IoT devices. What do you plan to do during the next reporting period to accomplish the goals? 1) We will conduct security experiments to study the efficiency of moving target defense for a smart farming system. We will develop an MDP framework to model the moving target defense scenario to improve the resiliency of smart farming system against diverse cyber-attacks. Then, we will design a robust algorithm using the deep reinforcement learning technique. We will also conduct simulation experiments to validate the effectiveness of the proposed DRL algorithms. We will compare the performance of the proposed DRL algorithm with other algorithms such as Proximal Policy Optimization and Trust Region Policy Optimization. 2) We will build a prototype to measure the performance of the smart farming system in a real setting. The prototype will involve a cloud and fog infrastructure, IoT gateways, IoT nodes, and an end-user device. Various smart farming applications such as soil management, yield prediction will be implemented and deployed on the cloud-fog infrastructure. These applications provide farmers with the real-time status of the farm, critical alerts, and useful insights to make informed farming decisions. We will evaluate the prototype using performance metrics such as response time, cost, and resource utilization. 3) We will reach out to farmers and exercise online questionnaire to receive data on types of crops, and challenges they face in managing their crop, soil, and water resources. After we identify the local farms, we will test the prototype and collect our experimental results. We will hold information sessions for the farmers to brief them about the usefulness of the IoT based smart farming system in providing a real-time update on the health of their farms, and providing analytics support for making better decisions.
Impacts
What was accomplished under these goals?
Our research activities resulted in new mathematical models and algorithms for optimizing the cost and ensuring the quality of service required by the IoT based smart farming system. We developed an IoT application model that enables the farmer to manage soil efficiently by incorporating functions such as soil monitoring, alert generation on detecting abnormal soil conditions, and soil analytics. We also obtained an in-depth understanding of the cyber threats that exist in smart farming system and significant insights on new methods that can be used to prevent the threats. This project provided research experiences to five undergraduate computer science students and helped them enhance their creativity and analytical skills. Moreover, the student researchers gained hands-on skill by writing the source code for the proposed algorithms using Python, and conducting numerous experiments using simulators. Goals Accomplished: This year, we accomplished the goals (1), (2), (5), and (6). The activities, experiments, and key outcomes are described as follows: Goal (1): We developed a mathematical model of the IoT gateway placement problem and IoT application placement problem using ILP. Our objective was to optimize the cost of deploying IoT gateways (also referred to as fog servers) in a farm area to meet the latency requirement of smart farming applications. As we still need a remote cloud server to run resource-intensive smart farming applications, we optimize the cost of deploying smart farming applications in a hybrid cloud-fog infrastructure. We designed several QoS constraints such as latency and security in our ILP formulation. Goal (2): To solve the IoT gateway placement problem, we designed placement algorithms using K-means clustering and Hedonic games. and Genetic Algorithms. The modified K-means clustering group the sensors deployed in a farm area into K clusters. The clusters are formed in way that reduces the energy consumption, and ensures effective resource utilization within a cluster. The cluster center represents the location of IoT gateway. The algorithm based on hedonic game aims at finding coalitions of sensors that minimize the energy cost and ensures quality of service (QoS). We also designed an IoT application placement algorithm using genetic algorithm (GA). The proposed GA based algorithm aims at determining the location (e.g., cloud server, fog server) for running the smart farming applications with reduced cost and better QoS. Goal (5): We implemented the ILP models of the IoT gateway placement and IoT application placement in IBM ILOG CPLEX Optimization Studio to study the optimal solution. We conducted multiple experiments by varying the number of sensors, the number of applications, and the QoS requirements (security and latency). We obtained results for various performance metrics such as cost, and resource utilization. Our results show that the cost increases with an increase in number of applications with stringent security and latency requirements. This is because fog servers are ideal candidates for hosting those applications, but require higher processing cost compared to the cloud server. One of our key observation is that fog servers experience same usage as cloud when 50% of the smart farming applications have stringent requirements. Goal (6): We conducted extensive experiments using the iFogSim simulator to evaluate the performance of different placement schemes: cloud-only, and fog-only. These schemes will serve as baseline schemes for evaluating the performance of our proposed algorithms. We use a smart farming scenario that includes a cloud server, one proxy-server (tier-2 fog server), and three gateways (tier-1 fog servers). We considered a distributed data flow (DDF) model of a soil management application that receives soil parameters from sensors, aggregate them, and provides users with real-time status of soil as well as critical alerts on detecting abnormal soil condition such as low moisture. The application also includes a lightweight analytic module that provides users with recommendations for optimal usage of water. We performed experiments to study the impact of the number of sensors on end-to-end latency, and network usage. We observe that the "fog-only" scheme outperforms the "cloud-based" scheme in terms of both latency and network usage. It shows the capabilities of fog servers in running latency-sensitive smart farming applications.
Publications
- Type:
Journal Articles
Status:
Submitted
Year Published:
2021
Citation:
J. Sahoo, K. Barrett, Internet of Things (IoT) Application Model for Smart Farming , Available at https://arxiv.org/ftp/arxiv/papers/2101/2101.03722.pdf submitted to IEEE SoutheastCon 2021.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2020
Citation:
IoT & Cybersecurity , Maritime Resilience and Cybersecurity Panel, 11th Maritime Risk Symposium, October, 2020 (Invited talk)
(Organized by the Critical Infrastructure Resilience Institute (CIRI), University of Illinois at Urbana-Champaign in collaboration with the National Academy of Sciences)
Video Link: https://www.youtube.com/watch?v=NbaaPEl4seg&t=3431s
|
Progress 08/16/19 to 09/30/19
Outputs
Target Audience: Three undergraduate students (Computer science major) participated in conducting the research activities of the project. The students were involved in surveying the literatures, developing mathematical frameworks and designing novel algorithms. Through their active participation, the students developed understanding of advanced computer science topics such as Internet-of-Things, cloud and fog computing. They also developed the strong analytical skills and the ability to design innovative algorithms. Moreover, the students also developed the ability to work effectively in a team as they often engaged in brain-storming sessions to discuss research problem and potential solutions. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?The undergraduate student researchers received training from the PI in identifying research problems, conducting comprehensive survey, conducting evaluations, and developing algorithms. These skills will enable the students to perform intensive and high-quality research. This project provided three undergraduate student researchers the opportunity to present the research problem, evaluation results of their literature survey, and the salient features of the algorithms (K-means, game theory, and Social Network Inspired Betweenness Centrality) they worked on this year. The presentation was held in the 1890 conference room in Fall 2019. Through the process, the students learned how to make a technical presentation and also developed their oral communication skills. How have the results been disseminated to communities of interest?
Nothing Reported
What do you plan to do during the next reporting period to accomplish the goals?We plan to conduct the following activities in the next reporting period to accomplish goals (3)-(6): We will conduct an in-depth survey of the related works that use moving target defense technique to ensure security of IoT devices. We will also evaluate the related works based on the key requirements such as overall resiliency, number of parameters randomized, etc. We will develop a MDP framework to model the MTD scenario. We will design a robust algorithm using deep reinforcement learning technique. We will conduct extensive simulations to validate the effectiveness of the proposed algorithms. We will also implement selected state-of-the-art for comparing their performance with the proposed algorithms. We will also conduct simulation experiments to validate the effectiveness of the proposed DRL algorithms. We will compare the performance of the proposed DRL algorithm with other algorithms such as Proximal Policy Optimization and Trust Region Policy Optimization.
Impacts
What was accomplished under these goals?
Change in Knowledge: This year, the undergraduate students obtained a deeper understanding of the mathematical frameworks which further strengthened their ability in building complex mathematical models and frameworks. A more detailed survey of the state-of-the-art on placement issues in IoT systems provided the research team with insights on new challenges and potential research directions. The study allowed the team to design new and innovative algorithms to overcome the challenges. In addition, the undergraduate student researchers received ample opportunity to collaborate with each other to achieve a common goal i.e. optimizing the cost and performance of the smart farming system. Change in Action: This year, the undergraduate students honed their analytical skills, critical thinking and creativity by actively involved in designing innovative algorithms. Goals Accomplished: We started working on Goal 1 and Goal 2 in Fall 2019 and we expect to accomplish them by Spring 2020. The detail activities for Goal 1 and Goal 2 that we undertook this year are described as follows: Goal 1: To achieve goal 1, first we identified the optimization objectives and constraints for two key problems in the smart farming scenario: IoT gateway placement and IoT Application Placement. We studied these two problems in more depth and determined that both problems require minimization of deployment cost and sensor energy cost and a constraint of Quality of Service (QoS) Guarantee. Minimizing deployment cost ensures a cost-effective placement of the gateways; whereas minimizing energy costs ensures energy-efficiency of the deployed sensors as they have limited on-board battery. The QoS guarantee ensures that the IoT applications provide results to the farmers in real-time. Our next step is to complete the optimization models for IoT gateway placement and IoT application placement using Integer Linear Programming (ILP). Goal 2: In order to design IoT gateway placement algorithms, we studied different approaches such as K-means, and game theory in more depth. Our next step is to design the algorithms in a way that they minimize the deployment cost, sensor energy cost and ensures QoS is satisfied. We also performed a comprehensive survey of existing algorithms that focus on placement issues in IoT domains. Our evaluation shows that none of the studied algorithms satisfy the requirements. The evaluation also shows that there is a need to design new and efficient algorithms. In order to design IoT application placement algorithm, we analyzed a social network analysis metric, called Betweenness Centrality. Our next step is to update the metric to fit it into the smart farming scenario.
Publications
|
|