Intelligent Algorithms And Their Applications On The Problems Of Fault Diagnosis In The Oil Field

Taking problems of fault diagnosis in the process of gathering and transportation of oil and gas as background, some research is made on some intelligent algorithms such as the immune algorithm, ant colony algorithm, wavelet analysis, neural network, and chaos theory. Then all these intelligent algorithms are applied in the leakage location for long distance crude oil transportation pipeline and fault diagnosis of pump-jack in the oil field. The ideas and the methods of the paper provide a new approach to solve fault diagnosis. The details are as follows:Using density regulating mechanism, individual diversity retaining strategy, and immune memory function, a self adaptive immune algorithm based on the niche isolation mechanism is put forward. Adopting the strategy of adaptive selection probability based on the density and fitness, the phenomenon of premature convergence of sample genetic algorithm is overcomed efficiently, which can not only enhance the similarity of the antibodies, but also can give attention to the diversity of the antibodies. Thus it can be ensured that the solution of the algorithm can not get into local optimum solutions and the searching time can be shortened. The algorithm is used to leakage location for long distance crude oil transportation pipeline, the accuracy of location is improved.According to the rules of ants looking for food and ideas of immune genetic, an ant colony genetic algorithm is proposed based on immune mechanism. In precondition of keeping characteristics of the ant colony algorithm for global optimization and getting rationality of solutions in finite time, by introducing immune mechanism, the diversity of individuals can be assured and the phenomenon of long searching time and easily stagnating are avoided.An adaptive multi-dimensional diagonal wavelet neural network based on single dilation compact wavelet frame is constructed after comparing many wavelet neural network structures and decomposing data using wavelet pack to get characteristics. The network includes the initial network and sub-networks adaptively incorporated during the training according to the accuracy. Through analysis of synthesis time domain of input signal in each dimension, choose the proper dilation parameter and displacement parameter to construct wavelet frame in every scale and form the sub-networks in each level, so that the problem of the"dimension disaster" in the multi-dimension wavelet networks can be solved. Aiming at the diagonal recurrent structure, derive out dynamic recusive least square with forget factor to train parameters of the network. The result of fault diagnosis of pump-jack shows that the fault diagnosis method based on wavelet neural network has much more efficiency and accuracy than the traditional BP network fault diagnosis method.On the base of analyzing chaotic property of immune mechanism, chaotic immune network is proposed. There are many characteristics of artificial immune algorithm, such as solving the diversity problem of the individual that is to be chosen, learning to memorize and searching in parallel efficiently. By using all these advantages, the structure and weights of the neural network are optimized, convergent velocity is improved, phenomenon of premature convergence is cleared up, the global optimization is realized. Using ergordicity and sensitivity of initial value on chaos system, the data redundancy is reduced, and hunting zone is enlarged. Accordingly, population diversity can be retained and local optimum can be avoided. The result of fault diagnosis of pump-jack shows that the ability of global optimization is very strong.A new negative selection algorithm is proposed combining the advantage of genetic algorithm and simulating annealing algorithm. The initial values of the detectors are initialized by genetic algorithm, thus the diversity of the detectors is retained, the scope of detecting is enlarged. The variable radius of detectors is introduced to cover non-self space efficiently. The redundancy of detectors is reduced and the efficency is improved by using simulating annealing.