The invention discloses a method for searching a gas leakage source by dynamically integrating a plurality of search strategies. The method comprises a stage of estimating adaptability of various search methods under different search conditions by means of an artificial neural network to determine a main factor which affects performance of various searching methods, and train the artificial neural network for estimating the adaptability of the search methods; an integration mode of a plume finding phase, namely, an integration strategy which combines the search method based on a visual attention mechanism with a random search method; an integration mode of a plume tracing phase, namely, respectively determining a search direction of the next step by a chemotaxis method, an anemotaxis method and a visual method when a robot detects the gas concentration information, and inputting the current search condition which is the main factor affecting the performance of the search methods to the trained artificial neural network to obtain the adaptability of various methods, and planning a local search route of the robot by a driving force vector composition method. The method can obviously improve the efficiency and the success rate of searching the gas leakage source.