The Alarming Rate Of Increase Of Blackouts Around The Globe Show That These Events Are No Longer As Rare As People Assumed And More Research And Development E®Ort Is Needed To Identify And Prevent Such Catastrophic Failures. Thus, There Is A Need To Develop Decision Support Tools For Identi¯Cation And Mitigation Of The Risk Of Such Events. This Thesis Focuses On The Development Of Such Tools For Identi¯Cation And Prevention Of Catastrophic Failures In Power Systems. In The Last Few Years, A Lot Of Research Has Been Carried Out To Analyze Such Events. Lack Of Computational Resources And E±Cient Algorithms Are The Major Obstacles For Studying Large Blackouts. In This Thesis An Approach For On-Line Identi¯Cation Of High Risk Cascades (Leading To Catastrophic Failures) Is Proposed. Identi¯Cation Process Starts With A Single Event From A Normal Operating Condition. To Generate The Normal Operating Condition A Multi Objective Optimal Power Flow (MOOPF) Method Based On Particle Swarm Optimization Is Developed. To Evaluate The Contingencies Accurately, Fast Dynamics Such As Transient Stability And E®Ect Of Automatic Controllers Such As Load Characteristics, Generator Free Governor Mode Of Operation, On- Load Transformer Tap Adjustment, Etc. Are Considered In The Simulation Of Contingencies. Identi¯Cation Of Cascades By Simulation Of Contingencies Takes Considerable Computa- Tional Time And Is Not Suitable For On-Line Applications. For On-Line Identi¯Cation Of Cascades A Pattern Recognition And Fuzzy Estimation Scheme Is Proposed. A Cost-E±Cient Congestion Management Method Is Developed As One Of The Preventive Tools For Slow Cascading Events. Two Objectives, Congestion And Cost Are Simultaneously Minimized Using The Multi Objective Particle Swarm Optimization (MOPSO). Gener- Ation Rescheduling Of Participating Generators Is Done To Overcome The Congestion. In Case Generation Rescheduling Alone Is Not Su±Cient To Alleviate Overload Then Load Shed- Ding Is Carried Out As A Last Resort For The Participating Loads. Simulation Results For A Number Of Standard IEEE Test Systems And One Indian System Are Presented To Show The E®Ectiveness Of The Proposed Method.