A Real-time Stability Prediction Model Based On Spark In Chemical Plant Installations

The petrochemical industry has always been China's pillar industry,some well-known chemical companies such as Sinopec,Chinese petroleum,Chinese CNOOC company has a huge number of fixed assets and a large number of practitioners.They represent the lifeblood of the GDP economy.All kinds of chemical products are widely used in traffic,construction,agriculture and other industrial fields,and they also appear in all aspects of daily life.The chemical industry has not only promoted the development of all walks of life,but also moisted the people all over the world.Chemical equipment is the floorboard of all the equipment used in the production of chemical products.The safe operation of chemical deviceis very significant.It is a necessary condition to ensure the sustained and stable chemical reaction.It is also an important prerequisite to protect people's life,property and company assets.However,the chemical industry has always belonged to high-risk industries,and all kinds of chemical accidents occur frequently,and most of the unexpected chemical accidents are closely linked to the chemical devices.The malfunction or damage of the chemical equipment will lead to the production delay,cause the project schedule delay and the economic loss,and it will cause the explosion,fire,or gas leak seriously,and leads to the serious casualties unfortunately.Therefore,ensuring the safe and stable operation of chemical equipment has become the primary task of all major chemical enterprises in the world.Companies have stepped up their awareness and actively studied risk assessment methods for chemical plants.Based on the big data analysis technology and frames,we built a real-time forecast model of chemical plant stability.The model takes the spark data analysis technique into chemical plants stability monitoring first time.The model of real-time chemical plants stability monitoring has the following highlights:(1)For the model of training:according to the data of chemical plant site push,custom design a set of effective "data cleaning",a set of rules "to" rule;in order to reduce the size of data,we design a set of "characteristic screening" rule,this rule can effectively save space and operation time,and provide a very efficient and accurate prediction effect.(2)For real-time data forecast part:according to the actual demand of chemical enterprises,proposes a "distributed message queue Kafka + spark streaming stream processing program" system architecture for the first time in the energy sector;chemical analysis technology,special design and optimization of a variety of big data,finally reasonable and successful application.(3)For the integration of model training part and real-time prediction part:in order to ensure that the training process is parallel with the real-time prediction process,the AKKA model is designed,so that the update and acquisition of the model results can be effectively isolated.Under the line,the algorithm model is trained with historical data,and the model results are stored in the model factory,and the model results are used directly on the line.The two processes do not interfere with each other.(4)For the actual prediction situation:because of the different actual prediction conditions of each chemical plant,this paper customize different models of training and prediction methods,and make use of different algorithms and strategies to carry out precise training for a large number of target locations.(5)For the results show:a custom designed analysis system,will forecast the actual situation and data comparison,draw the predicted map,given device stability conclusion,user view;in addition,the user can manually specify the system parameters,and all kinds of spark streaming issued the implementation of the rules of procedure instruction.A complete set of real-time prediction model for the stability of chemical plant based on spark has been operated in the Yanshan petrochemical plant.