Research On Gas Pollution Source Localization Algorithm Based On HMM And Reinforcement Learning

Due to the state’s investment in the physical manufacturing industry,the scale of the petrochemical industry has grown rapidly,and the expansion of the industrial scale has led to an increasing frequency of production,utilization and storage of hazardous chemicals.As hazardous chemicals are stored in highly polluting chemicals such as chlorine,ammonia and hydrogen sulfide,if the polluting gas leaks,it will cause significant damage to society and the environment.In order to reduce the continuous harm caused by the leakage of polluted gas and prepare for the follow-up rescue work,it is necessary to have the ability to accurately and quickly locate the source of gas pollution when an accident occurs.Traditional gas pollution source localization methods have problems such as complex design,low localization accuracy,long time consumption,and easy to fall into extreme value areas,which are difficult to put into use in real scenarios.This paper conducts in-depth research on the above problems,and proposes a gas pollution source localization algorithm to solve the pollution source localization problem of mobile robots in turbulent environment.The main research contents of this paper are as follows:（1）In order to adapt to the real environment,we first conduct modeling research on the gas diffusion state,transform the pollution source diffusion process into a Hidden Markov Model（HMM）,and use the forward derivation algorithm to calculate the pollution source at different times.The plume distribution map can effectively reduce the time complexity of the algorithm.Then,according to the plume distribution map,the pollutant gas release process is reversely solved,and a pollution source probability distribution map（source probability map）with lower computational complexity is derived.Fluent software was used to simulate the continuous release of ammonia gas as a polluting gas,and Matlab software was used to simulate the plume distribution map and source probability map.The results show that during the traversal process of the robot,the drawn pollution source probability map gradually shrinks within the real value range,and the plume distribution map is also close to the real value,which proves the reliability of the proposed method.（2）In order to improve the positioning accuracy of gas pollution sources and avoid falling into the extreme value region of the plume,the pollution source positioning problem is transformed into a reinforcement learning（RL）problem.In the reinforcement learning model establishment stage,the elements in the model are transformed into a framework suitable for pollution source localization;in the decision-making stage,the reward function is optimized,and the plume distribution map and the pollution source probability map are fused by fuzzy reasoning,and a proposal to reflect the two is proposed.The reward function for the weighted relationship.A fuzzy rule library is established to express the fuzzy relationship between the plume concentration value and the non-detection period of the plume and the weight coefficient of the reward function.The value of the weight coefficient of the reward function is solved by solving the fuzzy function,so as to guide the robot to explore the plume or track the target pollution source..Finally,the optimal path with the greatest value is solved through the value iteration algorithm,and the pollution source location problem is completed.Compared with the traditional pollution source localization experiments,the method in this paper can effectively improve the gas pollution source localization accuracy and shorten the localization time,indicating that the proposed method has good accuracy and effectiveness.