【正文】 in the boiler in order to monitor the water level. Whenever the water level is out of limits, the system will give an alarm. The boiler will not be allowed to work if its boiler drum water level control system is not intact.Since 2002, there have been several accidents happened in heatengine plants in our country that were caused by inadvertent maloperations of boiler drum water level control system. These accidents have caused many serious economic losses. Besides, It was reported that about 25% of emergency shutdowns in the nuclear power plants based on pressurizedwater reactor (PWR) are caused by poor control of the steam generator (SG) water level[1].We can expect that more and more problems that would be caused by insufficient accuracy of water level monitoring system or inplete actions of water level control system during the initiate of boiler will happen with the growing of thermal power generating units. Giving an overview of those major accidents happened in these years, we can find that the inaccuracy of measuring system and inappropriate signal processing are the main reasons.The traditional twooutofthree method for boiler drum water level control can be achieved by DCS modules, with a probability that the two signals selected are far from the true water level while the third one is much closer. Thus, a misjudgment is possible to happen. To improve, lately a Multimodel predictive control method[2], as well as a Fuzzy Logic method[3], has been used for nuclear steam generator. This method is useful but it still cannot increase the water level measuring accuracy. Because of this, only by various collections of information of the object and prehensive utilization of the information can we measure the water level more accurately and more effectively, and then can we increase the amplitude of control signal during the protection of boiler drum water level. With development of multisensor data fusion technology, the multisource data fusion technique——as an intelligent data processing technology, can eliminate uncertainties of the system and provide accurate measuring results and prehensive information[4]. This technology provides a new way to solve the plex uncertain problems in boiler water level protection. We use the neural network technology in this article.If there are too many characteristic parameters for artificial neural network to deal with, they will make the Artificial Neural Network Architecture too big and make the generalization ability of the network too weak and then lower the accuracy, thereby affecting the reliability of ANN. As many important problems in science and industry have been addressed by data mining methods[5], and inspired by this, we build an integrated neural network relating to signal types and an ensemble of feature vectors based on fuzzy theory and data fusion technique. We have used this technology in protection of boiler drum water level and it has raised the accuracy effectively pared with traditional protection system. With the fuzzy neural network technology, test data from a nuclearpower plant is used for approve the approach which presented in the paper. From the results, we can decrease the possibility of this kind of misclassification of the system and thus increase the accuracy of actions of alarm. boiler drum water level control system based on data fusion technologyFirstly, there should be many multitype gages when measuring boiler water level in order to ensure the accuracy of measurements. Generally, there should be at least three differentialpressure water gages for every unit. The measurement results would be used for the boiler drum water level control system which is based on data fusion technology and presented in this article.Secondly, in order to use the data fusion technology for boiler drum water level control system, the fuzzy processing of measurement results of these three water gages is necessary. With the fuzzy processing of measurement results, we can prehensively utilize all the information that es from these three water gages by neural network technology, rather than part of the information that we did with the traditional method—twooutofthree .At last, we will get the result after using the neural network technology into the boiler drum water level control system.The boiler drum water level control system based on data fusion technology are shown in fig1. Figure 1. The boiler drum water level control system based on data fusion Technology.III Implementation methods for data fusion A FuzzingIt’s necessary to set suitable membership function so as to fuzz data that e from water gages. The membership function is the base for the application of fuzzy set, and it is also the key to use ambiguity function the membership function is important, there isn’t any effective and universal way to set suitable membership function. Nowadays, the most mon way is to set by experience and then correct it according to experiments or simulations. In the measuring of boiler drum water level, people can use the following three membership functions if it is hard to find a suitable one[7]:? triangle membership function.? trapezoidal membership function? normal membership functionAnd among all the NNs, the BPNN, to whom special attentions should be paid[9], is widely used, because the mapping function and its allorder derivatives of any nonlinear system can be approached by a three layer feedforward neural network according to the accuracy we need. Thus, a three layer feedforward neural network can be used to represent any system no matter the mapping function of the system could be found or not.Though the BPNN is made up of simple neurons, it has a strong power to deal with nonlinear problems because of its network structure and algorithm. Here is an example of three layer feedforward neural network with just one neuron in the output layer. It vividly depicts the theory and process of BP network algorithm[10].The structure of three layer feedforward neural network with just o