The arising smart home, which is based on the Internet of things (IoT) technology, is facing serious information security risks, such as denial of service, illegal access or privacy disclosure. Existing solutions have problems in handling large amounts of high-dimensional data, such as low detection rate and high false detection one. A hybrid neural network intrusion detection method for smart home is proposed, which is to establish data features through deep learning, map high-dimensional data to low-dimensional data, and analyze and distinguish attack categories based on fuzzy neural network. A method for optimizing network depth is also provided, which can overcome the problems of the traditional method of determining the network layers' number based on experience. Therefore, it can not only ensure the detection accuracy, but also reduce time complexity of the algorithm. Test results prove that the detection rate of denial of service (DoS) attacks and remote illegal access can reach 94%, and the detection rate of some new attacks from the network can exceed 60% based on the proposed method.